tests: tune test_sharding_autosplit

.dockerignore

@@ -17,7 +17,6 @@
 !libs/
 !neon_local/
 !pageserver/
-!patches/
 !pgxn/
 !proxy/
 !s3_scrubber/

.github/actions/neon-branch-create/action.yml

@@ -3,13 +3,13 @@ description: 'Create Branch using API'
 
 inputs:
   api_key:
-    description: 'Neon API key'
+    desctiption: 'Neon API key'
     required: true
   project_id:
-    description: 'ID of the Project to create Branch in'
+    desctiption: 'ID of the Project to create Branch in'
     required: true
   api_host:
-    description: 'Neon API host'
+    desctiption: 'Neon API host'
     default: console-stage.neon.build
 outputs:
   dsn:

.github/actions/neon-branch-delete/action.yml

@@ -3,16 +3,16 @@ description: 'Delete Branch using API'
 
 inputs:
   api_key:
-    description: 'Neon API key'
+    desctiption: 'Neon API key'
     required: true
   project_id:
-    description: 'ID of the Project which should be deleted'
+    desctiption: 'ID of the Project which should be deleted'
     required: true
   branch_id:
-    description: 'ID of the branch to delete'
+    desctiption: 'ID of the branch to delete'
     required: true
   api_host:
-    description: 'Neon API host'
+    desctiption: 'Neon API host'
     default: console-stage.neon.build
 
 runs:

.github/actions/neon-project-create/action.yml

@@ -3,22 +3,22 @@ description: 'Create Neon Project using API'
 
 inputs:
   api_key:
-    description: 'Neon API key'
+    desctiption: 'Neon API key'
     required: true
   region_id:
-    description: 'Region ID, if not set the project will be created in the default region'
+    desctiption: 'Region ID, if not set the project will be created in the default region'
     default: aws-us-east-2
   postgres_version:
-    description: 'Postgres version; default is 15'
-    default: '15'
+    desctiption: 'Postgres version; default is 15'
+    default: 15
   api_host:
-    description: 'Neon API host'
+    desctiption: 'Neon API host'
     default: console-stage.neon.build
   provisioner:
-    description: 'k8s-pod or k8s-neonvm'
+    desctiption: 'k8s-pod or k8s-neonvm'
     default: 'k8s-pod'
   compute_units:
-    description: '[Min, Max] compute units; Min and Max are used for k8s-neonvm with autoscaling, for k8s-pod values Min and Max should be equal'
+    desctiption: '[Min, Max] compute units; Min and Max are used for k8s-neonvm with autoscaling, for k8s-pod values Min and Max should be equal'
     default: '[1, 1]'
 
 outputs:

.github/actions/neon-project-delete/action.yml

@@ -3,13 +3,13 @@ description: 'Delete Neon Project using API'
 
 inputs:
   api_key:
-    description: 'Neon API key'
+    desctiption: 'Neon API key'
     required: true
   project_id:
-    description: 'ID of the Project to delete'
+    desctiption: 'ID of the Project to delete'
     required: true
   api_host:
-    description: 'Neon API host'
+    desctiption: 'Neon API host'
     default: console-stage.neon.build
 
 runs:

.github/workflows/build_and_test.yml

@@ -548,7 +548,7 @@ jobs:
 
   report-benchmarks-failures:
     needs: [ benchmarks, create-test-report ]
-    if: github.ref_name == 'main' && failure() && needs.benchmarks.result == 'failure'
+    if: github.ref_name == 'main' && needs.benchmarks.result == 'failure'
     runs-on: ubuntu-latest
 
     steps:
@@ -723,13 +723,9 @@ jobs:
     uses: ./.github/workflows/trigger-e2e-tests.yml
     secrets: inherit
 
-  neon-image-arch:
+  neon-image:
     needs: [ check-permissions, build-build-tools-image, tag ]
-    strategy:
-      matrix:
-        arch: [ x64, arm64 ]
-
-    runs-on: ${{ fromJson(format('["self-hosted", "gen3", "{0}"]', matrix.arch == 'arm64' && 'large-arm64' || 'large')) }}
+    runs-on: [ self-hosted, gen3, large ]
 
     steps:
       - name: Checkout
@@ -751,6 +747,12 @@ jobs:
           username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
           password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
 
+      - uses: docker/login-action@v3
+        with:
+          registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
+          username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
+          password: ${{ secrets.AWS_SECRET_KEY_DEV }}
+
       - uses: docker/build-push-action@v5
         with:
           context: .
@@ -762,52 +764,25 @@ jobs:
           push: true
           pull: true
           file: Dockerfile
-          cache-from: type=registry,ref=neondatabase/neon:cache-${{ matrix.arch }}
-          cache-to: type=registry,ref=neondatabase/neon:cache-${{ matrix.arch }},mode=max
+          cache-from: type=registry,ref=neondatabase/neon:cache
+          cache-to: type=registry,ref=neondatabase/neon:cache,mode=max
           tags: |
-            neondatabase/neon:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}
+            369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:${{needs.tag.outputs.build-tag}}
+            neondatabase/neon:${{needs.tag.outputs.build-tag}}
 
       - name: Remove custom docker config directory
         if: always()
         run: |
           rm -rf .docker-custom
 
-  neon-image:
-    needs: [ neon-image-arch, tag ]
-    runs-on: ubuntu-latest
-
-    steps:
-      - uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
-          password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
-
-      - name: Create multi-arch image
-        run: |
-          docker buildx imagetools create -t neondatabase/neon:${{ needs.tag.outputs.build-tag }} \
-                                             neondatabase/neon:${{ needs.tag.outputs.build-tag }}-x64 \
-                                             neondatabase/neon:${{ needs.tag.outputs.build-tag }}-arm64
-
-      - uses: docker/login-action@v3
-        with:
-          registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
-          username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          password: ${{ secrets.AWS_SECRET_KEY_DEV }}
-
-      - name: Push multi-arch image to ECR
-        run: |
-          docker buildx imagetools create -t 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:${{ needs.tag.outputs.build-tag }} \
-                                                                                neondatabase/neon:${{ needs.tag.outputs.build-tag }}
-
-  compute-node-image-arch:
+  compute-node-image:
     needs: [ check-permissions, build-build-tools-image, tag ]
+    runs-on: [ self-hosted, gen3, large ]
+
     strategy:
       fail-fast: false
       matrix:
         version: [ v14, v15, v16 ]
-        arch: [ x64, arm64 ]
-
-    runs-on: ${{ fromJson(format('["self-hosted", "gen3", "{0}"]', matrix.arch == 'arm64' && 'large-arm64' || 'large')) }}
 
     steps:
       - name: Checkout
@@ -854,14 +829,15 @@ jobs:
           push: true
           pull: true
           file: Dockerfile.compute-node
-          cache-from: type=registry,ref=neondatabase/compute-node-${{ matrix.version }}:cache-${{ matrix.arch }}
-          cache-to: type=registry,ref=neondatabase/compute-node-${{ matrix.version }}:cache-${{ matrix.arch }},mode=max
+          cache-from: type=registry,ref=neondatabase/compute-node-${{ matrix.version }}:cache
+          cache-to: type=registry,ref=neondatabase/compute-node-${{ matrix.version }}:cache,mode=max
           tags: |
-            neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}
+            369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
+            neondatabase/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
 
       - name: Build compute-tools image
         # compute-tools are Postgres independent, so build it only once
-        if: matrix.version == 'v16'
+        if: ${{ matrix.version == 'v16' }}
         uses: docker/build-push-action@v5
         with:
           target: compute-tools-image
@@ -875,57 +851,14 @@ jobs:
           pull: true
           file: Dockerfile.compute-node
           tags: |
-            neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}-${{ matrix.arch }}
+            369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{ needs.tag.outputs.build-tag }}
+            neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}
 
       - name: Remove custom docker config directory
         if: always()
         run: |
           rm -rf .docker-custom
 
-  compute-node-image:
-    needs: [ compute-node-image-arch, tag ]
-    runs-on: ubuntu-latest
-
-    strategy:
-      matrix:
-        version: [ v14, v15, v16 ]
-
-    steps:
-      - uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
-          password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
-
-      - name: Create multi-arch compute-node image
-        run: |
-          docker buildx imagetools create -t neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }} \
-                                             neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-x64 \
-                                             neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}-arm64
-
-      - name: Create multi-arch compute-tools image
-        if: matrix.version == 'v16'
-        run: |
-          docker buildx imagetools create -t neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }} \
-                                             neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}-x64 \
-                                             neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}-arm64
-
-      - uses: docker/login-action@v3
-        with:
-          registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
-          username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          password: ${{ secrets.AWS_SECRET_KEY_DEV }}
-
-      - name: Push multi-arch compute-node-${{ matrix.version }} image to ECR
-        run: |
-          docker buildx imagetools create -t 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }} \
-                                                                                neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}
-
-      - name: Push multi-arch compute-tools image to ECR
-        if: matrix.version == 'v16'
-        run: |
-          docker buildx imagetools create -t 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{ needs.tag.outputs.build-tag }} \
-                                                                                neondatabase/compute-tools:${{ needs.tag.outputs.build-tag }}
-
   vm-compute-node-image:
     needs: [ check-permissions, tag, compute-node-image ]
     runs-on: [ self-hosted, gen3, large ]
@@ -933,8 +866,11 @@ jobs:
       fail-fast: false
       matrix:
         version: [ v14, v15, v16 ]
+    defaults:
+      run:
+        shell: sh -eu {0}
     env:
-      VM_BUILDER_VERSION: v0.29.3
+      VM_BUILDER_VERSION: v0.28.1
 
     steps:
       - name: Checkout
@@ -947,48 +883,26 @@ jobs:
           curl -fL https://github.com/neondatabase/autoscaling/releases/download/$VM_BUILDER_VERSION/vm-builder -o vm-builder
           chmod +x vm-builder
 
-      # Use custom DOCKER_CONFIG directory to avoid conflicts with default settings
-      # The default value is ~/.docker
-      - name: Set custom docker config directory
-        run: |
-          mkdir -p .docker-custom
-          echo DOCKER_CONFIG=$(pwd)/.docker-custom >> $GITHUB_ENV
-
-      - uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
-          password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
-
       # Note: we need a separate pull step here because otherwise vm-builder will try to pull, and
       # it won't have the proper authentication (written at v0.6.0)
       - name: Pulling compute-node image
         run: |
-          docker pull neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}
+          docker pull 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
 
       - name: Build vm image
         run: |
           ./vm-builder \
             -spec=vm-image-spec.yaml \
-            -src=neondatabase/compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }} \
-            -dst=neondatabase/vm-compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}
+            -src=369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}} \
+            -dst=369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
 
       - name: Pushing vm-compute-node image
         run: |
-          docker push neondatabase/vm-compute-node-${{ matrix.version }}:${{ needs.tag.outputs.build-tag }}
-
-      - name: Remove custom docker config directory
-        if: always()
-        run: |
-          rm -rf .docker-custom
+          docker push 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-${{ matrix.version }}:${{needs.tag.outputs.build-tag}}
 
   test-images:
     needs: [ check-permissions, tag, neon-image, compute-node-image ]
-    strategy:
-      fail-fast: false
-      matrix:
-        arch: [ x64, arm64 ]
-
-    runs-on: ${{ fromJson(format('["self-hosted", "gen3", "{0}"]', matrix.arch == 'arm64' && 'small-arm64' || 'small')) }}
+    runs-on: [ self-hosted, gen3, small ]
 
     steps:
       - name: Checkout
@@ -1006,7 +920,7 @@ jobs:
       - name: Verify image versions
         shell: bash # ensure no set -e for better error messages
         run: |
-          pageserver_version=$(docker run --rm neondatabase/neon:${{ needs.tag.outputs.build-tag }} "/bin/sh" "-c" "/usr/local/bin/pageserver --version")
+          pageserver_version=$(docker run --rm 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:${{needs.tag.outputs.build-tag}} "/bin/sh" "-c" "/usr/local/bin/pageserver --version")
 
           echo "Pageserver version string: $pageserver_version"
 
@@ -1032,48 +946,78 @@ jobs:
 
   promote-images:
     needs: [ check-permissions, tag, test-images, vm-compute-node-image ]
-    runs-on: ubuntu-latest
-
-    env:
-      VERSIONS: v14 v15 v16
+    runs-on: [ self-hosted, gen3, small ]
+    container: golang:1.19-bullseye
+    # Don't add if-condition here.
+    # The job should always be run because we have dependant other jobs that shouldn't be skipped
 
     steps:
-      - uses: docker/login-action@v3
-        with:
-          username: ${{ secrets.NEON_DOCKERHUB_USERNAME }}
-          password: ${{ secrets.NEON_DOCKERHUB_PASSWORD }}
-
-      - uses: docker/login-action@v3
-        with:
-          registry: 369495373322.dkr.ecr.eu-central-1.amazonaws.com
-          username: ${{ secrets.AWS_ACCESS_KEY_DEV }}
-          password: ${{ secrets.AWS_SECRET_KEY_DEV }}
-
-      - name: Copy vm-compute-node images to ECR
+      - name: Install Crane & ECR helper
         run: |
-          for version in ${VERSIONS}; do
-            docker buildx imagetools create -t 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-${version}:${{ needs.tag.outputs.build-tag }} \
-                                               neondatabase/vm-compute-node-${version}:${{ needs.tag.outputs.build-tag }}
-          done
+          go install github.com/google/go-containerregistry/cmd/crane@31786c6cbb82d6ec4fb8eb79cd9387905130534e # v0.11.0
+          go install github.com/awslabs/amazon-ecr-credential-helper/ecr-login/cli/docker-credential-ecr-login@69c85dc22db6511932bbf119e1a0cc5c90c69a7f # v0.6.0
+
+      - name: Configure ECR login
+        run: |
+          mkdir /github/home/.docker/
+          echo "{\"credsStore\":\"ecr-login\"}" > /github/home/.docker/config.json
+
+      - name: Copy vm-compute-node images to Docker Hub
+        run: |
+          crane pull 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v14:${{needs.tag.outputs.build-tag}} vm-compute-node-v14
+          crane pull 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v15:${{needs.tag.outputs.build-tag}} vm-compute-node-v15
+          crane pull 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v16:${{needs.tag.outputs.build-tag}} vm-compute-node-v16
 
       - name: Add latest tag to images
-        if: github.ref_name == 'main'
+        if: github.ref_name == 'main' || github.ref_name == 'release' || github.ref_name == 'release-proxy'
         run: |
-          for repo in neondatabase 369495373322.dkr.ecr.eu-central-1.amazonaws.com; do
-            docker buildx imagetools create -t $repo/neon:latest \
-                                               $repo/neon:${{ needs.tag.outputs.build-tag }}
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v14:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v14:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v15:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v15:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v16:${{needs.tag.outputs.build-tag}} latest
+          crane tag 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v16:${{needs.tag.outputs.build-tag}} latest
 
-            docker buildx imagetools create -t $repo/compute-tools:latest \
-                                               $repo/compute-tools:${{ needs.tag.outputs.build-tag }}
+      - name: Push images to production ECR
+        if: github.ref_name == 'main' || github.ref_name == 'release'|| github.ref_name == 'release-proxy'
+        run: |
+          crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/neon:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/neon:latest
+          crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/compute-tools:latest
+          crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v14:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v14:latest
+          crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v14:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v14:latest
+          crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v15:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v15:latest
+          crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v15:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v15:latest
+          crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v16:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/compute-node-v16:latest
+          crane copy 369495373322.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v16:${{needs.tag.outputs.build-tag}} 093970136003.dkr.ecr.eu-central-1.amazonaws.com/vm-compute-node-v16:latest
 
-            for version in ${VERSIONS}; do
-              docker buildx imagetools create -t $repo/compute-node-${version}:latest \
-                                                 $repo/compute-node-${version}:${{ needs.tag.outputs.build-tag }}
+      - name: Configure Docker Hub login
+        run: |
+          # ECR Credential Helper & Docker Hub don't work together in config, hence reset
+          echo "" > /github/home/.docker/config.json
+          crane auth login -u ${{ secrets.NEON_DOCKERHUB_USERNAME }} -p ${{ secrets.NEON_DOCKERHUB_PASSWORD }} index.docker.io
 
-              docker buildx imagetools create -t $repo/vm-compute-node-${version}:latest \
-                                                 $repo/vm-compute-node-${version}:${{ needs.tag.outputs.build-tag }}
-            done
-          done
+      - name: Push vm-compute-node to Docker Hub
+        run: |
+          crane push vm-compute-node-v14 neondatabase/vm-compute-node-v14:${{needs.tag.outputs.build-tag}}
+          crane push vm-compute-node-v15 neondatabase/vm-compute-node-v15:${{needs.tag.outputs.build-tag}}
+          crane push vm-compute-node-v16 neondatabase/vm-compute-node-v16:${{needs.tag.outputs.build-tag}}
+
+      - name: Push latest tags to Docker Hub
+        if: github.ref_name == 'main' || github.ref_name == 'release'|| github.ref_name == 'release-proxy'
+        run: |
+          crane tag neondatabase/neon:${{needs.tag.outputs.build-tag}} latest
+          crane tag neondatabase/compute-tools:${{needs.tag.outputs.build-tag}} latest
+          crane tag neondatabase/compute-node-v14:${{needs.tag.outputs.build-tag}} latest
+          crane tag neondatabase/vm-compute-node-v14:${{needs.tag.outputs.build-tag}} latest
+          crane tag neondatabase/compute-node-v15:${{needs.tag.outputs.build-tag}} latest
+          crane tag neondatabase/vm-compute-node-v15:${{needs.tag.outputs.build-tag}} latest
+          crane tag neondatabase/compute-node-v16:${{needs.tag.outputs.build-tag}} latest
+          crane tag neondatabase/vm-compute-node-v16:${{needs.tag.outputs.build-tag}} latest
+
+      - name: Cleanup ECR folder
+        run: rm -rf ~/.ecr
 
   trigger-custom-extensions-build-and-wait:
     needs: [ check-permissions, tag ]

Cargo.lock

@@ -708,7 +708,7 @@ dependencies = [
  "sha1",
  "sync_wrapper",
  "tokio",
- "tokio-tungstenite",
+ "tokio-tungstenite 0.20.0",
  "tower",
  "tower-layer",
  "tower-service",
@@ -979,12 +979,6 @@ version = "3.13.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "a3e2c3daef883ecc1b5d58c15adae93470a91d425f3532ba1695849656af3fc1"
 
-[[package]]
-name = "bytemuck"
-version = "1.16.0"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "78834c15cb5d5efe3452d58b1e8ba890dd62d21907f867f383358198e56ebca5"
-
 [[package]]
 name = "byteorder"
 version = "1.4.3"
@@ -1072,9 +1066,9 @@ dependencies = [
 
 [[package]]
 name = "chrono"
-version = "0.4.38"
+version = "0.4.31"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "a21f936df1771bf62b77f047b726c4625ff2e8aa607c01ec06e5a05bd8463401"
+checksum = "7f2c685bad3eb3d45a01354cedb7d5faa66194d1d58ba6e267a8de788f79db38"
 dependencies = [
  "android-tzdata",
  "iana-time-zone",
@@ -1082,7 +1076,7 @@ dependencies = [
  "num-traits",
  "serde",
  "wasm-bindgen",
- "windows-targets 0.52.4",
+ "windows-targets 0.48.0",
 ]
 
 [[package]]
@@ -1109,7 +1103,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "defaa24ecc093c77630e6c15e17c51f5e187bf35ee514f4e2d67baaa96dae22b"
 dependencies = [
  "ciborium-io",
- "half 1.8.2",
+ "half",
 ]
 
 [[package]]
@@ -1239,10 +1233,8 @@ dependencies = [
  "serde_json",
  "signal-hook",
  "tar",
- "thiserror",
  "tokio",
  "tokio-postgres",
- "tokio-stream",
  "tokio-util",
  "toml_edit",
  "tracing",
@@ -1471,21 +1463,26 @@ dependencies = [
 
 [[package]]
 name = "crossbeam-deque"
-version = "0.8.5"
+version = "0.8.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "613f8cc01fe9cf1a3eb3d7f488fd2fa8388403e97039e2f73692932e291a770d"
+checksum = "ce6fd6f855243022dcecf8702fef0c297d4338e226845fe067f6341ad9fa0cef"
 dependencies = [
+ "cfg-if",
  "crossbeam-epoch",
  "crossbeam-utils",
 ]
 
 [[package]]
 name = "crossbeam-epoch"
-version = "0.9.18"
+version = "0.9.14"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "5b82ac4a3c2ca9c3460964f020e1402edd5753411d7737aa39c3714ad1b5420e"
+checksum = "46bd5f3f85273295a9d14aedfb86f6aadbff6d8f5295c4a9edb08e819dcf5695"
 dependencies = [
+ "autocfg",
+ "cfg-if",
  "crossbeam-utils",
+ "memoffset 0.8.0",
+ "scopeguard",
 ]
 
 [[package]]
@@ -1599,7 +1596,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "6943ae99c34386c84a470c499d3414f66502a41340aa895406e0d2e4a207b91d"
 dependencies = [
  "cfg-if",
- "hashbrown 0.14.5",
+ "hashbrown 0.14.0",
  "lock_api",
  "once_cell",
  "parking_lot_core 0.9.8",
@@ -2000,27 +1997,6 @@ dependencies = [
  "percent-encoding",
 ]
 
-[[package]]
-name = "framed-websockets"
-version = "0.1.0"
-source = "git+https://github.com/neondatabase/framed-websockets#34eff3d6f8cfccbc5f35e4f65314ff7328621127"
-dependencies = [
- "base64 0.21.1",
- "bytemuck",
- "bytes",
- "futures-core",
- "futures-sink",
- "http-body-util",
- "hyper 1.2.0",
- "hyper-util",
- "pin-project",
- "rand 0.8.5",
- "sha1",
- "thiserror",
- "tokio",
- "tokio-util",
-]
-
 [[package]]
 name = "fs2"
 version = "0.4.3"
@@ -2273,17 +2249,6 @@ version = "1.8.2"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "eabb4a44450da02c90444cf74558da904edde8fb4e9035a9a6a4e15445af0bd7"
 
-[[package]]
-name = "half"
-version = "2.4.1"
-source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6dd08c532ae367adf81c312a4580bc67f1d0fe8bc9c460520283f4c0ff277888"
-dependencies = [
- "cfg-if",
- "crunchy",
- "num-traits",
-]
-
 [[package]]
 name = "hash32"
 version = "0.3.1"
@@ -2310,9 +2275,9 @@ dependencies = [
 
 [[package]]
 name = "hashbrown"
-version = "0.14.5"
+version = "0.14.0"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "e5274423e17b7c9fc20b6e7e208532f9b19825d82dfd615708b70edd83df41f1"
+checksum = "2c6201b9ff9fd90a5a3bac2e56a830d0caa509576f0e503818ee82c181b3437a"
 dependencies = [
  "ahash",
  "allocator-api2",
@@ -2320,11 +2285,11 @@ dependencies = [
 
 [[package]]
 name = "hashlink"
-version = "0.9.1"
+version = "0.8.4"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "6ba4ff7128dee98c7dc9794b6a411377e1404dba1c97deb8d1a55297bd25d8af"
+checksum = "e8094feaf31ff591f651a2664fb9cfd92bba7a60ce3197265e9482ebe753c8f7"
 dependencies = [
- "hashbrown 0.14.5",
+ "hashbrown 0.14.0",
 ]
 
 [[package]]
@@ -2633,6 +2598,21 @@ dependencies = [
  "tokio-native-tls",
 ]
 
+[[package]]
+name = "hyper-tungstenite"
+version = "0.13.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "7a343d17fe7885302ed7252767dc7bb83609a874b6ff581142241ec4b73957ad"
+dependencies = [
+ "http-body-util",
+ "hyper 1.2.0",
+ "hyper-util",
+ "pin-project-lite",
+ "tokio",
+ "tokio-tungstenite 0.21.0",
+ "tungstenite 0.21.0",
+]
+
 [[package]]
 name = "hyper-util"
 version = "0.1.3"
@@ -2710,7 +2690,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "ad227c3af19d4914570ad36d30409928b75967c298feb9ea1969db3a610bb14e"
 dependencies = [
  "equivalent",
- "hashbrown 0.14.5",
+ "hashbrown 0.14.0",
 ]
 
 [[package]]
@@ -2972,7 +2952,7 @@ version = "0.12.3"
 source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "d3262e75e648fce39813cb56ac41f3c3e3f65217ebf3844d818d1f9398cfb0dc"
 dependencies = [
- "hashbrown 0.14.5",
+ "hashbrown 0.14.0",
 ]
 
 [[package]]
@@ -3025,7 +3005,7 @@ checksum = "652bc741286361c06de8cb4d89b21a6437f120c508c51713663589eeb9928ac5"
 dependencies = [
  "bytes",
  "crossbeam-utils",
- "hashbrown 0.14.5",
+ "hashbrown 0.14.0",
  "itoa",
  "lasso",
  "measured-derive",
@@ -3587,7 +3567,7 @@ source = "registry+https://github.com/rust-lang/crates.io-index"
 checksum = "49203cdcae0030493bad186b28da2fa25645fa276a51b6fec8010d281e02ef79"
 dependencies = [
  "dlv-list",
- "hashbrown 0.14.5",
+ "hashbrown 0.14.0",
 ]
 
 [[package]]
@@ -3908,14 +3888,13 @@ dependencies = [
 
 [[package]]
 name = "parquet"
-version = "51.0.0"
-source = "git+https://github.com/apache/arrow-rs?branch=master#2534976a564be3d2d56312dc88fb1b6ed4cef829"
+version = "49.0.0"
+source = "git+https://github.com/neondatabase/arrow-rs?branch=neon-fix-bugs#8a0bc58aa67b98aabbd8eee7c6ca4281967ff9e9"
 dependencies = [
  "ahash",
  "bytes",
  "chrono",
- "half 2.4.1",
- "hashbrown 0.14.5",
+ "hashbrown 0.14.0",
  "num",
  "num-bigint",
  "paste",
@@ -3923,13 +3902,12 @@ dependencies = [
  "thrift",
  "twox-hash",
  "zstd",
- "zstd-sys",
 ]
 
 [[package]]
 name = "parquet_derive"
-version = "51.0.0"
-source = "git+https://github.com/apache/arrow-rs?branch=master#2534976a564be3d2d56312dc88fb1b6ed4cef829"
+version = "49.0.0"
+source = "git+https://github.com/neondatabase/arrow-rs?branch=neon-fix-bugs#8a0bc58aa67b98aabbd8eee7c6ca4281967ff9e9"
 dependencies = [
  "parquet",
  "proc-macro2",
@@ -3956,9 +3934,9 @@ checksum = "de3145af08024dea9fa9914f381a17b8fc6034dfb00f3a84013f7ff43f29ed4c"
 
 [[package]]
 name = "pbkdf2"
-version = "0.12.2"
+version = "0.12.1"
 source = "registry+https://github.com/rust-lang/crates.io-index"
-checksum = "f8ed6a7761f76e3b9f92dfb0a60a6a6477c61024b775147ff0973a02653abaf2"
+checksum = "f0ca0b5a68607598bf3bad68f32227a8164f6254833f84eafaac409cd6746c31"
 dependencies = [
  "digest",
  "hmac",
@@ -4381,7 +4359,6 @@ dependencies = [
 name = "proxy"
 version = "0.1.0"
 dependencies = [
- "ahash",
  "anyhow",
  "async-compression",
  "async-trait",
@@ -4398,14 +4375,12 @@ dependencies = [
  "chrono",
  "clap",
  "consumption_metrics",
- "crossbeam-deque",
  "dashmap",
  "env_logger",
  "fallible-iterator",
- "framed-websockets",
  "futures",
  "git-version",
- "hashbrown 0.14.5",
+ "hashbrown 0.13.2",
  "hashlink",
  "hex",
  "hmac",
@@ -4415,6 +4390,7 @@ dependencies = [
  "humantime",
  "hyper 0.14.26",
  "hyper 1.2.0",
+ "hyper-tungstenite",
  "hyper-util",
  "indexmap 2.0.1",
  "ipnet",
@@ -4459,6 +4435,7 @@ dependencies = [
  "smol_str",
  "socket2 0.5.5",
  "subtle",
+ "sync_wrapper",
  "task-local-extensions",
  "thiserror",
  "tikv-jemalloc-ctl",
@@ -4467,7 +4444,6 @@ dependencies = [
  "tokio-postgres",
  "tokio-postgres-rustls",
  "tokio-rustls 0.25.0",
- "tokio-tungstenite",
  "tokio-util",
  "tower-service",
  "tracing",
@@ -6404,7 +6380,19 @@ dependencies = [
  "futures-util",
  "log",
  "tokio",
- "tungstenite",
+ "tungstenite 0.20.1",
+]
+
+[[package]]
+name = "tokio-tungstenite"
+version = "0.21.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "c83b561d025642014097b66e6c1bb422783339e0909e4429cde4749d1990bc38"
+dependencies = [
+ "futures-util",
+ "log",
+ "tokio",
+ "tungstenite 0.21.0",
 ]
 
 [[package]]
@@ -6418,7 +6406,7 @@ dependencies = [
  "futures-io",
  "futures-sink",
  "futures-util",
- "hashbrown 0.14.5",
+ "hashbrown 0.14.0",
  "pin-project-lite",
  "tokio",
  "tracing",
@@ -6700,6 +6688,25 @@ dependencies = [
  "utf-8",
 ]
 
+[[package]]
+name = "tungstenite"
+version = "0.21.0"
+source = "registry+https://github.com/rust-lang/crates.io-index"
+checksum = "9ef1a641ea34f399a848dea702823bbecfb4c486f911735368f1f137cb8257e1"
+dependencies = [
+ "byteorder",
+ "bytes",
+ "data-encoding",
+ "http 1.1.0",
+ "httparse",
+ "log",
+ "rand 0.8.5",
+ "sha1",
+ "thiserror",
+ "url",
+ "utf-8",
+]
+
 [[package]]
 name = "twox-hash"
 version = "1.6.3"
@@ -7470,7 +7477,6 @@ dependencies = [
 name = "workspace_hack"
 version = "0.1.0"
 dependencies = [
- "ahash",
  "anyhow",
  "aws-config",
  "aws-runtime",
@@ -7496,7 +7502,7 @@ dependencies = [
  "futures-sink",
  "futures-util",
  "getrandom 0.2.11",
- "hashbrown 0.14.5",
+ "hashbrown 0.14.0",
  "hex",
  "hmac",
  "hyper 0.14.26",

Cargo.toml

@@ -41,7 +41,6 @@ license = "Apache-2.0"
 
 ## All dependency versions, used in the project
 [workspace.dependencies]
-ahash = "0.8"
 anyhow = { version = "1.0", features = ["backtrace"] }
 arc-swap = "1.6"
 async-compression = { version = "0.4.0", features = ["tokio", "gzip", "zstd"] }
@@ -75,7 +74,6 @@ clap = { version = "4.0", features = ["derive"] }
 comfy-table = "6.1"
 const_format = "0.2"
 crc32c = "0.6"
-crossbeam-deque = "0.8.5"
 crossbeam-utils = "0.8.5"
 dashmap = { version = "5.5.0", features = ["raw-api"] }
 either = "1.8"
@@ -83,14 +81,13 @@ enum-map = "2.4.2"
 enumset = "1.0.12"
 fail = "0.5.0"
 fallible-iterator = "0.2"
-framed-websockets = { version = "0.1.0", git = "https://github.com/neondatabase/framed-websockets" }
 fs2 = "0.4.3"
 futures = "0.3"
 futures-core = "0.3"
 futures-util = "0.3"
 git-version = "0.3"
-hashbrown = "0.14"
-hashlink = "0.9.1"
+hashbrown = "0.13"
+hashlink = "0.8.4"
 hdrhistogram = "7.5.2"
 hex = "0.4"
 hex-literal = "0.4"
@@ -101,7 +98,7 @@ http-types = { version = "2", default-features = false }
 humantime = "2.1"
 humantime-serde = "1.1.1"
 hyper = "0.14"
-tokio-tungstenite = "0.20.0"
+hyper-tungstenite = "0.13.0"
 indexmap = "2"
 inotify = "0.10.2"
 ipnet = "2.9.0"
@@ -124,8 +121,8 @@ opentelemetry = "0.20.0"
 opentelemetry-otlp = { version = "0.13.0", default_features=false, features = ["http-proto", "trace", "http", "reqwest-client"] }
 opentelemetry-semantic-conventions = "0.12.0"
 parking_lot = "0.12"
-parquet = { version = "51.0.0", default-features = false, features = ["zstd"] }
-parquet_derive = "51.0.0"
+parquet = { version = "49.0.0", default-features = false, features = ["zstd"] }
+parquet_derive = "49.0.0"
 pbkdf2 = { version = "0.12.1", features = ["simple", "std"] }
 pin-project-lite = "0.2"
 procfs = "0.14"
@@ -246,8 +243,8 @@ tonic-build = "0.9"
 tokio-postgres = { git = "https://github.com/neondatabase/rust-postgres.git", branch="neon" }
 
 # bug fixes for UUID
-parquet = { git = "https://github.com/apache/arrow-rs", branch = "master" }
-parquet_derive = { git = "https://github.com/apache/arrow-rs", branch = "master" }
+parquet = { git = "https://github.com/neondatabase/arrow-rs", branch = "neon-fix-bugs" }
+parquet_derive = { git = "https://github.com/neondatabase/arrow-rs", branch = "neon-fix-bugs" }
 
 ################# Binary contents sections
 

Dockerfile.compute-node

@@ -241,17 +241,11 @@ RUN wget https://github.com/df7cb/postgresql-unit/archive/refs/tags/7.7.tar.gz -
 FROM build-deps AS vector-pg-build
 COPY --from=pg-build /usr/local/pgsql/ /usr/local/pgsql/
 
-COPY patches/pgvector.patch /pgvector.patch
-
-# By default, pgvector Makefile uses `-march=native`. We don't want that, 
-# because we build the images on different machines than where we run them.
-# Pass OPTFLAGS="" to remove it.
-RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.7.0.tar.gz -O pgvector.tar.gz && \
-    echo "1b5503a35c265408b6eb282621c5e1e75f7801afc04eecb950796cfee2e3d1d8 pgvector.tar.gz" | sha256sum --check && \
+RUN wget https://github.com/pgvector/pgvector/archive/refs/tags/v0.5.1.tar.gz -O pgvector.tar.gz && \
+    echo "cc7a8e034a96e30a819911ac79d32f6bc47bdd1aa2de4d7d4904e26b83209dc8 pgvector.tar.gz" | sha256sum --check && \
     mkdir pgvector-src && cd pgvector-src && tar xvzf ../pgvector.tar.gz --strip-components=1 -C . && \
-    patch -p1 < /pgvector.patch && \
-    make -j $(getconf _NPROCESSORS_ONLN) OPTFLAGS="" PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
-    make -j $(getconf _NPROCESSORS_ONLN) OPTFLAGS="" install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
+    make -j $(getconf _NPROCESSORS_ONLN) install PG_CONFIG=/usr/local/pgsql/bin/pg_config && \
     echo 'trusted = true' >> /usr/local/pgsql/share/extension/vector.control
 
 #########################################################################################

README.md

@@ -1,6 +1,4 @@
-[![Neon](https://github.com/neondatabase/neon/assets/11527560/f15a17f0-836e-40c5-b35d-030606a6b660)](https://neon.tech)
-
-
+[![Neon](https://user-images.githubusercontent.com/13738772/236813940-dcfdcb5b-69d3-449b-a686-013febe834d4.png)](https://neon.tech)
 
 # Neon
 

compute_tools/Cargo.toml

@@ -27,12 +27,10 @@ reqwest = { workspace = true, features = ["json"] }
 tokio = { workspace = true, features = ["rt", "rt-multi-thread"] }
 tokio-postgres.workspace = true
 tokio-util.workspace = true
-tokio-stream.workspace = true
 tracing.workspace = true
 tracing-opentelemetry.workspace = true
 tracing-subscriber.workspace = true
 tracing-utils.workspace = true
-thiserror.workspace = true
 url.workspace = true
 
 compute_api.workspace = true

compute_tools/src/catalog.rs

@@ -1,116 +0,0 @@
-use compute_api::{
-    responses::CatalogObjects,
-    spec::{Database, Role},
-};
-use futures::Stream;
-use postgres::{Client, NoTls};
-use std::{path::Path, process::Stdio, result::Result, sync::Arc};
-use tokio::{
-    io::{AsyncBufReadExt, BufReader},
-    process::Command,
-    task,
-};
-use tokio_stream::{self as stream, StreamExt};
-use tokio_util::codec::{BytesCodec, FramedRead};
-use tracing::warn;
-
-use crate::{
-    compute::ComputeNode,
-    pg_helpers::{get_existing_dbs, get_existing_roles},
-};
-
-pub async fn get_dbs_and_roles(compute: &Arc<ComputeNode>) -> anyhow::Result<CatalogObjects> {
-    let connstr = compute.connstr.clone();
-    task::spawn_blocking(move || {
-        let mut client = Client::connect(connstr.as_str(), NoTls)?;
-        let roles: Vec<Role>;
-        {
-            let mut xact = client.transaction()?;
-            roles = get_existing_roles(&mut xact)?;
-        }
-        let databases: Vec<Database> = get_existing_dbs(&mut client)?.values().cloned().collect();
-
-        Ok(CatalogObjects { roles, databases })
-    })
-    .await?
-}
-
-#[derive(Debug, thiserror::Error)]
-pub enum SchemaDumpError {
-    #[error("Database does not exist.")]
-    DatabaseDoesNotExist,
-    #[error("Failed to execute pg_dump.")]
-    IO(#[from] std::io::Error),
-}
-
-// It uses the pg_dump utility to dump the schema of the specified database.
-// The output is streamed back to the caller and supposed to be streamed via HTTP.
-//
-// Before return the result with the output, it checks that pg_dump produced any output.
-// If not, it tries to parse the stderr output to determine if the database does not exist
-// and special error is returned.
-//
-// To make sure that the process is killed when the caller drops the stream, we use tokio kill_on_drop feature.
-pub async fn get_database_schema(
-    compute: &Arc<ComputeNode>,
-    dbname: &str,
-) -> Result<impl Stream<Item = Result<bytes::Bytes, std::io::Error>>, SchemaDumpError> {
-    let pgbin = &compute.pgbin;
-    let basepath = Path::new(pgbin).parent().unwrap();
-    let pgdump = basepath.join("pg_dump");
-    let mut connstr = compute.connstr.clone();
-    connstr.set_path(dbname);
-    let mut cmd = Command::new(pgdump)
-        .arg("--schema-only")
-        .arg(connstr.as_str())
-        .stdout(Stdio::piped())
-        .stderr(Stdio::piped())
-        .kill_on_drop(true)
-        .spawn()?;
-
-    let stdout = cmd.stdout.take().ok_or_else(|| {
-        std::io::Error::new(std::io::ErrorKind::Other, "Failed to capture stdout.")
-    })?;
-
-    let stderr = cmd.stderr.take().ok_or_else(|| {
-        std::io::Error::new(std::io::ErrorKind::Other, "Failed to capture stderr.")
-    })?;
-
-    let mut stdout_reader = FramedRead::new(stdout, BytesCodec::new());
-    let stderr_reader = BufReader::new(stderr);
-
-    let first_chunk = match stdout_reader.next().await {
-        Some(Ok(bytes)) if !bytes.is_empty() => bytes,
-        Some(Err(e)) => {
-            return Err(SchemaDumpError::IO(e));
-        }
-        _ => {
-            let mut lines = stderr_reader.lines();
-            if let Some(line) = lines.next_line().await? {
-                if line.contains(&format!("FATAL:  database \"{}\" does not exist", dbname)) {
-                    return Err(SchemaDumpError::DatabaseDoesNotExist);
-                }
-                warn!("pg_dump stderr: {}", line)
-            }
-            tokio::spawn(async move {
-                while let Ok(Some(line)) = lines.next_line().await {
-                    warn!("pg_dump stderr: {}", line)
-                }
-            });
-
-            return Err(SchemaDumpError::IO(std::io::Error::new(
-                std::io::ErrorKind::Other,
-                "failed to start pg_dump",
-            )));
-        }
-    };
-    let initial_stream = stream::once(Ok(first_chunk.freeze()));
-    // Consume stderr and log warnings
-    tokio::spawn(async move {
-        let mut lines = stderr_reader.lines();
-        while let Ok(Some(line)) = lines.next_line().await {
-            warn!("pg_dump stderr: {}", line)
-        }
-    });
-    Ok(initial_stream.chain(stdout_reader.map(|res| res.map(|b| b.freeze()))))
-}

compute_tools/src/http/api.rs

@@ -5,21 +5,17 @@ use std::net::SocketAddr;
 use std::sync::Arc;
 use std::thread;
 
-use crate::catalog::SchemaDumpError;
-use crate::catalog::{get_database_schema, get_dbs_and_roles};
 use crate::compute::forward_termination_signal;
 use crate::compute::{ComputeNode, ComputeState, ParsedSpec};
 use compute_api::requests::ConfigurationRequest;
 use compute_api::responses::{ComputeStatus, ComputeStatusResponse, GenericAPIError};
 
 use anyhow::Result;
-use hyper::header::CONTENT_TYPE;
 use hyper::service::{make_service_fn, service_fn};
 use hyper::{Body, Method, Request, Response, Server, StatusCode};
 use tokio::task;
 use tracing::{error, info, warn};
 use tracing_utils::http::OtelName;
-use utils::http::request::must_get_query_param;
 
 fn status_response_from_state(state: &ComputeState) -> ComputeStatusResponse {
     ComputeStatusResponse {
@@ -137,34 +133,6 @@ async fn routes(req: Request<Body>, compute: &Arc<ComputeNode>) -> Response<Body
             }
         }
 
-        (&Method::GET, "/dbs_and_roles") => {
-            info!("serving /dbs_and_roles GET request",);
-            match get_dbs_and_roles(compute).await {
-                Ok(res) => render_json(Body::from(serde_json::to_string(&res).unwrap())),
-                Err(_) => {
-                    render_json_error("can't get dbs and roles", StatusCode::INTERNAL_SERVER_ERROR)
-                }
-            }
-        }
-
-        (&Method::GET, "/database_schema") => {
-            let database = match must_get_query_param(&req, "database") {
-                Err(e) => return e.into_response(),
-                Ok(database) => database,
-            };
-            info!("serving /database_schema GET request with database: {database}",);
-            match get_database_schema(compute, &database).await {
-                Ok(res) => render_plain(Body::wrap_stream(res)),
-                Err(SchemaDumpError::DatabaseDoesNotExist) => {
-                    render_json_error("database does not exist", StatusCode::NOT_FOUND)
-                }
-                Err(e) => {
-                    error!("can't get schema dump: {}", e);
-                    render_json_error("can't get schema dump", StatusCode::INTERNAL_SERVER_ERROR)
-                }
-            }
-        }
-
         // download extension files from remote extension storage on demand
         (&Method::POST, route) if route.starts_with("/extension_server/") => {
             info!("serving {:?} POST request", route);
@@ -335,25 +303,10 @@ fn render_json_error(e: &str, status: StatusCode) -> Response<Body> {
     };
     Response::builder()
         .status(status)
-        .header(CONTENT_TYPE, "application/json")
         .body(Body::from(serde_json::to_string(&error).unwrap()))
         .unwrap()
 }
 
-fn render_json(body: Body) -> Response<Body> {
-    Response::builder()
-        .header(CONTENT_TYPE, "application/json")
-        .body(body)
-        .unwrap()
-}
-
-fn render_plain(body: Body) -> Response<Body> {
-    Response::builder()
-        .header(CONTENT_TYPE, "text/plain")
-        .body(body)
-        .unwrap()
-}
-
 async fn handle_terminate_request(compute: &Arc<ComputeNode>) -> Result<(), (String, StatusCode)> {
     {
         let mut state = compute.state.lock().unwrap();

compute_tools/src/http/openapi_spec.yaml

@@ -68,51 +68,6 @@ paths:
               schema:
                 $ref: "#/components/schemas/Info"
 
-  /dbs_and_roles:
-    get:
-      tags:
-        - Info
-      summary: Get databases and roles in the catalog.
-      description: ""
-      operationId: getDbsAndRoles
-      responses:
-        200:
-          description: Compute schema objects
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/DbsAndRoles"
-
-  /database_schema:
-    get:
-      tags:
-        - Info
-      summary: Get schema dump
-      parameters:
-        - name: database
-          in: query
-          description: Database name to dump.
-          required: true
-          schema:
-            type: string
-          example: "postgres"
-      description: Get schema dump in SQL format.
-      operationId: getDatabaseSchema
-      responses:
-        200:
-          description: Schema dump
-          content:
-            text/plain:
-              schema:
-                type: string
-                description: Schema dump in SQL format.
-        404:
-          description: Non existing database.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/GenericError"
-
   /check_writability:
     post:
       tags:
@@ -274,73 +229,6 @@ components:
         num_cpus:
           type: integer
 
-    DbsAndRoles:
-      type: object
-      description: Databases and Roles
-      required:
-        - roles
-        - databases
-      properties:
-        roles:
-          type: array
-          items:
-            $ref: "#/components/schemas/Role"
-        databases:
-          type: array
-          items:
-            $ref: "#/components/schemas/Database"
-
-    Database:
-      type: object
-      description: Database
-      required:
-        - name
-        - owner
-        - restrict_conn
-        - invalid
-      properties:
-        name:
-          type: string
-        owner:
-          type: string
-        options:
-          type: array
-          items:
-            $ref: "#/components/schemas/GenericOption"
-        restrict_conn:
-          type: boolean
-        invalid:
-          type: boolean
-
-    Role:
-      type: object
-      description: Role
-      required:
-        - name
-      properties:
-        name:
-          type: string
-        encrypted_password:
-          type: string
-        options:
-          type: array
-          items:
-            $ref: "#/components/schemas/GenericOption"
-
-    GenericOption:
-      type: object
-      description: Schema Generic option
-      required:
-        - name
-        - vartype
-      properties:
-        name:
-          type: string
-        value:
-          type: string
-        vartype:
-          type: string
-
     ComputeState:
       type: object
       required:

compute_tools/src/lib.rs

@@ -8,7 +8,6 @@ pub mod configurator;
 pub mod http;
 #[macro_use]
 pub mod logger;
-pub mod catalog;
 pub mod compute;
 pub mod extension_server;
 pub mod monitor;

compute_tools/src/swap.rs

@@ -1,5 +1,3 @@
-use std::path::Path;
-
 use anyhow::{anyhow, Context};
 use tracing::warn;
 
@@ -19,24 +17,17 @@ pub fn resize_swap(size_bytes: u64) -> anyhow::Result<()> {
         .arg(size_bytes.to_string())
         .spawn();
 
+    if matches!(&child_result, Err(e) if e.kind() == std::io::ErrorKind::NotFound) {
+        warn!("ignoring \"not found\" error from resize-swap to avoid swapoff while compute is running");
+        return Ok(());
+    }
+
     child_result
         .context("spawn() failed")
         .and_then(|mut child| child.wait().context("wait() failed"))
         .and_then(|status| match status.success() {
             true => Ok(()),
-            false => {
-                // The command failed. Maybe it was because the resize-swap file doesn't exist?
-                // The --once flag causes it to delete itself on success so we don't disable swap
-                // while postgres is running; maybe this is fine.
-                match Path::new(RESIZE_SWAP_BIN).try_exists() {
-                    Err(_) | Ok(true) => Err(anyhow!("process exited with {status}")),
-                    // The path doesn't exist; we're actually ok 
-                    Ok(false) => {
-                        warn!("ignoring \"not found\" error from resize-swap to avoid swapoff while compute is running");
-                        Ok(())
-                    },
-                }
-            }
+            false => Err(anyhow!("process exited with {status}")),
         })
         // wrap any prior error with the overall context that we couldn't run the command
         .with_context(|| {

control_plane/src/storage_controller.rs

@@ -243,13 +243,9 @@ impl StorageController {
                 anyhow::bail!("initdb failed with status {status}");
             }
 
-            // Write a minimal config file:
-            // - Specify the port, since this is chosen dynamically
-            // - Switch off fsync, since we're running on lightweight test environments and when e.g. scale testing
-            //   the storage controller we don't want a slow local disk to interfere with that.
             tokio::fs::write(
                 &pg_data_path.join("postgresql.conf"),
-                format!("port = {}\nfsync=off\n", self.postgres_port),
+                format!("port = {}", self.postgres_port),
             )
             .await?;
         };

docker-compose/compute_wrapper/Dockerfile

@@ -1,4 +1,4 @@
-ARG REPOSITORY=neondatabase
+ARG REPOSITORY=369495373322.dkr.ecr.eu-central-1.amazonaws.com
 ARG COMPUTE_IMAGE=compute-node-v14
 ARG TAG=latest
 

docker-compose/docker_compose_test.sh

@@ -8,6 +8,8 @@
 # Their defaults point at DockerHub `neondatabase/neon:latest` image.`,
 # to verify custom image builds (e.g pre-published ones).
 
+# XXX: Current does not work on M1 macs due to x86_64 Docker images compiled only, and no seccomp support in M1 Docker emulation layer.
+
 set -eux -o pipefail
 
 SCRIPT_DIR="$( cd -- "$( dirname -- "${BASH_SOURCE[0]}" )" &> /dev/null && pwd )"

libs/compute_api/src/responses.rs

@@ -3,7 +3,7 @@
 use chrono::{DateTime, Utc};
 use serde::{Deserialize, Serialize, Serializer};
 
-use crate::spec::{ComputeSpec, Database, Role};
+use crate::spec::ComputeSpec;
 
 #[derive(Serialize, Debug, Deserialize)]
 pub struct GenericAPIError {
@@ -113,12 +113,6 @@ pub struct ComputeMetrics {
     pub total_ext_download_size: u64,
 }
 
-#[derive(Clone, Debug, Default, Serialize)]
-pub struct CatalogObjects {
-    pub roles: Vec<Role>,
-    pub databases: Vec<Database>,
-}
-
 /// Response of the `/computes/{compute_id}/spec` control-plane API.
 /// This is not actually a compute API response, so consider moving
 /// to a different place.

libs/pageserver_api/src/keyspace.rs

@@ -307,7 +307,7 @@ impl KeySpace {
     }
 
     /// Merge another keyspace into the current one.
-    /// Note: the keyspaces must not overlap (enforced via assertions). To merge overlapping key ranges, use `KeySpaceRandomAccum`.
+    /// Note: the keyspaces must not ovelap (enforced via assertions)
     pub fn merge(&mut self, other: &KeySpace) {
         let all_ranges = self
             .ranges

libs/pageserver_api/src/models.rs

@@ -9,7 +9,7 @@ use std::{
     collections::HashMap,
     io::{BufRead, Read},
     num::{NonZeroU64, NonZeroUsize},
-    sync::atomic::AtomicUsize,
+    str::FromStr,
     time::{Duration, SystemTime},
 };
 
@@ -161,22 +161,6 @@ impl std::fmt::Debug for TenantState {
     }
 }
 
-/// A temporary lease to a specific lsn inside a timeline.
-/// Access to the lsn is guaranteed by the pageserver until the expiration indicated by `valid_until`.
-#[serde_as]
-#[derive(Debug, Clone, serde::Serialize, serde::Deserialize)]
-pub struct LsnLease {
-    #[serde_as(as = "SystemTimeAsRfc3339Millis")]
-    pub valid_until: SystemTime,
-}
-
-serde_with::serde_conv!(
-    SystemTimeAsRfc3339Millis,
-    SystemTime,
-    |time: &SystemTime| humantime::format_rfc3339_millis(*time).to_string(),
-    |value: String| -> Result<_, humantime::TimestampError> { humantime::parse_rfc3339(&value) }
-);
-
 /// The only [`TenantState`] variants we could be `TenantState::Activating` from.
 #[derive(Clone, Copy, Debug, PartialEq, Eq, serde::Serialize, serde::Deserialize)]
 pub enum ActivatingFrom {
@@ -305,7 +289,7 @@ pub struct TenantConfig {
     pub compaction_period: Option<String>,
     pub compaction_threshold: Option<usize>,
     // defer parsing compaction_algorithm, like eviction_policy
-    pub compaction_algorithm: Option<CompactionAlgorithmSettings>,
+    pub compaction_algorithm: Option<CompactionAlgorithm>,
     pub gc_horizon: Option<u64>,
     pub gc_period: Option<String>,
     pub image_creation_threshold: Option<usize>,
@@ -324,100 +308,28 @@ pub struct TenantConfig {
     pub switch_aux_file_policy: Option<AuxFilePolicy>,
 }
 
-/// The policy for the aux file storage. It can be switched through `switch_aux_file_policy`
-/// tenant config. When the first aux file written, the policy will be persisted in the
-/// `index_part.json` file and has a limited migration path.
-///
-/// Currently, we only allow the following migration path:
-///
-/// Unset -> V1
-///       -> V2
-///       -> CrossValidation -> V2
-#[derive(
-    Eq,
-    PartialEq,
-    Debug,
-    Copy,
-    Clone,
-    strum_macros::EnumString,
-    strum_macros::Display,
-    serde_with::DeserializeFromStr,
-    serde_with::SerializeDisplay,
-)]
-#[strum(serialize_all = "kebab-case")]
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
 pub enum AuxFilePolicy {
-    /// V1 aux file policy: store everything in AUX_FILE_KEY
-    #[strum(ascii_case_insensitive)]
     V1,
-    /// V2 aux file policy: store in the AUX_FILE keyspace
-    #[strum(ascii_case_insensitive)]
     V2,
-    /// Cross validation runs both formats on the write path and does validation
-    /// on the read path.
-    #[strum(ascii_case_insensitive)]
     CrossValidation,
 }
 
-impl AuxFilePolicy {
-    pub fn is_valid_migration_path(from: Option<Self>, to: Self) -> bool {
-        matches!(
-            (from, to),
-            (None, _) | (Some(AuxFilePolicy::CrossValidation), AuxFilePolicy::V2)
-        )
-    }
+impl FromStr for AuxFilePolicy {
+    type Err = anyhow::Error;
 
-    /// If a tenant writes aux files without setting `switch_aux_policy`, this value will be used.
-    pub fn default_tenant_config() -> Self {
-        Self::V1
-    }
-}
-
-/// The aux file policy memory flag. Users can store `Option<AuxFilePolicy>` into this atomic flag. 0 == unspecified.
-pub struct AtomicAuxFilePolicy(AtomicUsize);
-
-impl AtomicAuxFilePolicy {
-    pub fn new(policy: Option<AuxFilePolicy>) -> Self {
-        Self(AtomicUsize::new(
-            policy.map(AuxFilePolicy::to_usize).unwrap_or_default(),
-        ))
-    }
-
-    pub fn load(&self) -> Option<AuxFilePolicy> {
-        match self.0.load(std::sync::atomic::Ordering::Acquire) {
-            0 => None,
-            other => Some(AuxFilePolicy::from_usize(other)),
+    fn from_str(s: &str) -> Result<Self, Self::Err> {
+        let s = s.to_lowercase();
+        if s == "v1" {
+            Ok(Self::V1)
+        } else if s == "v2" {
+            Ok(Self::V2)
+        } else if s == "crossvalidation" || s == "cross_validation" {
+            Ok(Self::CrossValidation)
+        } else {
+            anyhow::bail!("cannot parse {} to aux file policy", s)
         }
     }
-
-    pub fn store(&self, policy: Option<AuxFilePolicy>) {
-        self.0.store(
-            policy.map(AuxFilePolicy::to_usize).unwrap_or_default(),
-            std::sync::atomic::Ordering::Release,
-        );
-    }
-}
-
-impl AuxFilePolicy {
-    pub fn to_usize(self) -> usize {
-        match self {
-            Self::V1 => 1,
-            Self::CrossValidation => 2,
-            Self::V2 => 3,
-        }
-    }
-
-    pub fn try_from_usize(this: usize) -> Option<Self> {
-        match this {
-            1 => Some(Self::V1),
-            2 => Some(Self::CrossValidation),
-            3 => Some(Self::V2),
-            _ => None,
-        }
-    }
-
-    pub fn from_usize(this: usize) -> Self {
-        Self::try_from_usize(this).unwrap()
-    }
 }
 
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
@@ -438,28 +350,13 @@ impl EvictionPolicy {
     }
 }
 
-#[derive(
-    Eq,
-    PartialEq,
-    Debug,
-    Copy,
-    Clone,
-    strum_macros::EnumString,
-    strum_macros::Display,
-    serde_with::DeserializeFromStr,
-    serde_with::SerializeDisplay,
-)]
-#[strum(serialize_all = "kebab-case")]
+#[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
+#[serde(tag = "kind")]
 pub enum CompactionAlgorithm {
     Legacy,
     Tiered,
 }
 
-#[derive(Eq, PartialEq, Debug, Clone, Serialize, Deserialize)]
-pub struct CompactionAlgorithmSettings {
-    pub kind: CompactionAlgorithm,
-}
-
 #[derive(Debug, Clone, Copy, PartialEq, Eq, Serialize, Deserialize)]
 pub struct EvictionPolicyLayerAccessThreshold {
     #[serde(with = "humantime_serde")]
@@ -707,9 +604,6 @@ pub struct TimelineInfo {
     pub state: TimelineState,
 
     pub walreceiver_status: String,
-
-    /// The last aux file policy being used on this timeline
-    pub last_aux_file_policy: Option<AuxFilePolicy>,
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
@@ -816,8 +710,6 @@ pub enum HistoricLayerInfo {
         lsn_end: Lsn,
         remote: bool,
         access_stats: LayerAccessStats,
-
-        l0: bool,
     },
     Image {
         layer_file_name: String,
@@ -870,16 +762,6 @@ pub struct DownloadRemoteLayersTaskSpawnRequest {
     pub max_concurrent_downloads: NonZeroUsize,
 }
 
-#[derive(Debug, Serialize, Deserialize)]
-pub struct IngestAuxFilesRequest {
-    pub aux_files: HashMap<String, String>,
-}
-
-#[derive(Debug, Serialize, Deserialize)]
-pub struct ListAuxFilesRequest {
-    pub lsn: Lsn,
-}
-
 #[derive(Debug, Serialize, Deserialize, Clone)]
 pub struct DownloadRemoteLayersTaskInfo {
     pub task_id: String,
@@ -1416,7 +1298,6 @@ impl PagestreamBeMessage {
 #[cfg(test)]
 mod tests {
     use serde_json::json;
-    use std::str::FromStr;
 
     use super::*;
 
@@ -1624,69 +1505,4 @@ mod tests {
             assert_eq!(actual, expected, "example on {line}");
         }
     }
-
-    #[test]
-    fn test_aux_file_migration_path() {
-        assert!(AuxFilePolicy::is_valid_migration_path(
-            None,
-            AuxFilePolicy::V1
-        ));
-        assert!(AuxFilePolicy::is_valid_migration_path(
-            None,
-            AuxFilePolicy::V2
-        ));
-        assert!(AuxFilePolicy::is_valid_migration_path(
-            None,
-            AuxFilePolicy::CrossValidation
-        ));
-        // Self-migration is not a valid migration path, and the caller should handle it by itself.
-        assert!(!AuxFilePolicy::is_valid_migration_path(
-            Some(AuxFilePolicy::V1),
-            AuxFilePolicy::V1
-        ));
-        assert!(!AuxFilePolicy::is_valid_migration_path(
-            Some(AuxFilePolicy::V2),
-            AuxFilePolicy::V2
-        ));
-        assert!(!AuxFilePolicy::is_valid_migration_path(
-            Some(AuxFilePolicy::CrossValidation),
-            AuxFilePolicy::CrossValidation
-        ));
-        // Migrations not allowed
-        assert!(!AuxFilePolicy::is_valid_migration_path(
-            Some(AuxFilePolicy::CrossValidation),
-            AuxFilePolicy::V1
-        ));
-        assert!(!AuxFilePolicy::is_valid_migration_path(
-            Some(AuxFilePolicy::V1),
-            AuxFilePolicy::V2
-        ));
-        assert!(!AuxFilePolicy::is_valid_migration_path(
-            Some(AuxFilePolicy::V2),
-            AuxFilePolicy::V1
-        ));
-        assert!(!AuxFilePolicy::is_valid_migration_path(
-            Some(AuxFilePolicy::V2),
-            AuxFilePolicy::CrossValidation
-        ));
-        assert!(!AuxFilePolicy::is_valid_migration_path(
-            Some(AuxFilePolicy::V1),
-            AuxFilePolicy::CrossValidation
-        ));
-        // Migrations allowed
-        assert!(AuxFilePolicy::is_valid_migration_path(
-            Some(AuxFilePolicy::CrossValidation),
-            AuxFilePolicy::V2
-        ));
-    }
-
-    #[test]
-    fn test_aux_parse() {
-        assert_eq!(AuxFilePolicy::from_str("V2").unwrap(), AuxFilePolicy::V2);
-        assert_eq!(AuxFilePolicy::from_str("v2").unwrap(), AuxFilePolicy::V2);
-        assert_eq!(
-            AuxFilePolicy::from_str("cross-validation").unwrap(),
-            AuxFilePolicy::CrossValidation
-        );
-    }
 }

libs/pageserver_api/src/shard.rs

@@ -559,14 +559,6 @@ impl ShardIdentity {
         }
     }
 
-    /// Obtains the shard number and count combined into a `ShardIndex`.
-    pub fn shard_index(&self) -> ShardIndex {
-        ShardIndex {
-            shard_count: self.count,
-            shard_number: self.number,
-        }
-    }
-
     pub fn shard_slug(&self) -> String {
         if self.count > ShardCount(0) {
             format!("-{:02x}{:02x}", self.number.0, self.count.0)

libs/postgres_backend/src/lib.rs

@@ -820,11 +820,10 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
         Ok(ProcessMsgResult::Continue)
     }
 
-    /// - Log as info/error result of handling COPY stream and send back
-    ///   ErrorResponse if that makes sense.
-    /// - Shutdown the stream if we got Terminate.
-    /// - Then close the connection because we don't handle exiting from COPY
-    ///   stream normally.
+    /// Log as info/error result of handling COPY stream and send back
+    /// ErrorResponse if that makes sense. Shutdown the stream if we got
+    /// Terminate. TODO: transition into waiting for Sync msg if we initiate the
+    /// close.
     pub async fn handle_copy_stream_end(&mut self, end: CopyStreamHandlerEnd) {
         use CopyStreamHandlerEnd::*;
 
@@ -850,6 +849,10 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
             }
         }
 
+        if let Terminate = &end {
+            self.state = ProtoState::Closed;
+        }
+
         let err_to_send_and_errcode = match &end {
             ServerInitiated(_) => Some((end.to_string(), SQLSTATE_SUCCESSFUL_COMPLETION)),
             Other(_) => Some((format!("{end:#}"), SQLSTATE_INTERNAL_ERROR)),
@@ -879,12 +882,6 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> PostgresBackend<IO> {
                 error!("failed to send ErrorResponse: {}", ee);
             }
         }
-
-        // Proper COPY stream finishing to continue using the connection is not
-        // implemented at the server side (we don't need it so far). To prevent
-        // further usages of the connection, close it.
-        self.framed.shutdown().await.ok();
-        self.state = ProtoState::Closed;
     }
 }
 

libs/postgres_connection/src/lib.rs

@@ -178,13 +178,6 @@ impl PgConnectionConfig {
     }
 }
 
-impl fmt::Display for PgConnectionConfig {
-    fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
-        // The password is intentionally hidden and not part of this display string.
-        write!(f, "postgresql://{}:{}", self.host, self.port)
-    }
-}
-
 impl fmt::Debug for PgConnectionConfig {
     fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result {
         // We want `password: Some(REDACTED-STRING)`, not `password: Some("REDACTED-STRING")`

libs/postgres_ffi/wal_craft/src/lib.rs

@@ -359,7 +359,7 @@ impl Crafter for LastWalRecordXlogSwitchEndsOnPageBoundary {
             // Is there enough space on the page for another logical message and an
             // XLOG_SWITCH? If not, start over.
             let page_remain = XLOG_BLCKSZ as u64 - u64::from(after_lsn) % XLOG_BLCKSZ as u64;
-            if page_remain < base_size + XLOG_SIZE_OF_XLOG_RECORD as u64 {
+            if page_remain < base_size - XLOG_SIZE_OF_XLOG_RECORD as u64 {
                 continue;
             }
 
@@ -373,29 +373,31 @@ impl Crafter for LastWalRecordXlogSwitchEndsOnPageBoundary {
                 "SELECT pg_logical_emit_message(false, 'swch', REPEAT('a', $1))",
                 &[&(repeats as i32)],
             )?;
-            info!(
-                "current_wal_insert_lsn={}, XLOG_SIZE_OF_XLOG_RECORD={}",
-                client.pg_current_wal_insert_lsn()?,
-                XLOG_SIZE_OF_XLOG_RECORD
-            );
-
-            // Emit the XLOG_SWITCH
-            let before_xlog_switch = client.pg_current_wal_insert_lsn()?;
-            let xlog_switch_record_end: PgLsn =
-                client.query_one("SELECT pg_switch_wal()", &[])?.get(0);
-
-            if u64::from(xlog_switch_record_end) as usize % XLOG_BLCKSZ
-                != XLOG_SIZE_OF_XLOG_SHORT_PHD
-            {
-                warn!(
-                    "XLOG_SWITCH message ended not on page boundary: {}, offset = {}, repeating",
-                    xlog_switch_record_end,
-                    u64::from(xlog_switch_record_end) as usize % XLOG_BLCKSZ
-                );
-                continue;
-            }
-            return Ok(vec![before_xlog_switch, xlog_switch_record_end]);
+            break;
         }
+        info!(
+            "current_wal_insert_lsn={}, XLOG_SIZE_OF_XLOG_RECORD={}",
+            client.pg_current_wal_insert_lsn()?,
+            XLOG_SIZE_OF_XLOG_RECORD
+        );
+
+        // Emit the XLOG_SWITCH
+        let before_xlog_switch = client.pg_current_wal_insert_lsn()?;
+        let xlog_switch_record_end: PgLsn = client.query_one("SELECT pg_switch_wal()", &[])?.get(0);
+        let next_segment = PgLsn::from(0x0200_0000);
+        ensure!(
+            xlog_switch_record_end < next_segment,
+            "XLOG_SWITCH record ended on or after the expected segment boundary: {} > {}",
+            xlog_switch_record_end,
+            next_segment
+        );
+        ensure!(
+            u64::from(xlog_switch_record_end) as usize % XLOG_BLCKSZ == XLOG_SIZE_OF_XLOG_SHORT_PHD,
+            "XLOG_SWITCH message ended not on page boundary: {}, offset = {}",
+            xlog_switch_record_end,
+            u64::from(xlog_switch_record_end) as usize % XLOG_BLCKSZ
+        );
+        Ok(vec![before_xlog_switch, xlog_switch_record_end])
     }
 }
 

libs/safekeeper_api/src/models.rs

@@ -50,9 +50,6 @@ pub struct SkTimelineInfo {
     pub safekeeper_connstr: Option<String>,
     #[serde(default)]
     pub http_connstr: Option<String>,
-    // Minimum of all active RO replicas flush LSN
-    #[serde(default = "lsn_invalid")]
-    pub standby_horizon: Lsn,
 }
 
 #[derive(Debug, Clone, Deserialize, Serialize)]

libs/walproposer/src/walproposer.rs

@@ -496,9 +496,9 @@ mod tests {
                 // TODO: When updating Postgres versions, this test will cause
                 // problems. Postgres version in message needs updating.
                 //
-                // Greeting(ProposerGreeting { protocol_version: 2, pg_version: 160003, proposer_id: [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], system_id: 0, timeline_id: 9e4c8f36063c6c6e93bc20d65a820f3d, tenant_id: 9e4c8f36063c6c6e93bc20d65a820f3d, tli: 1, wal_seg_size: 16777216 })
+                // Greeting(ProposerGreeting { protocol_version: 2, pg_version: 160002, proposer_id: [0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], system_id: 0, timeline_id: 9e4c8f36063c6c6e93bc20d65a820f3d, tenant_id: 9e4c8f36063c6c6e93bc20d65a820f3d, tli: 1, wal_seg_size: 16777216 })
                 vec![
-                    103, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 3, 113, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
+                    103, 0, 0, 0, 0, 0, 0, 0, 2, 0, 0, 0, 2, 113, 2, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
                     0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 158, 76, 143, 54, 6, 60, 108, 110,
                     147, 188, 32, 214, 90, 130, 15, 61, 158, 76, 143, 54, 6, 60, 108, 110, 147,
                     188, 32, 214, 90, 130, 15, 61, 1, 0, 0, 0, 0, 0, 0, 1,

pageserver/client/src/mgmt_api.rs

@@ -1,12 +1,8 @@
-use std::collections::HashMap;
-
-use bytes::Bytes;
 use pageserver_api::{models::*, shard::TenantShardId};
 use reqwest::{IntoUrl, Method, StatusCode};
 use utils::{
     http::error::HttpErrorBody,
     id::{TenantId, TimelineId},
-    lsn::Lsn,
 };
 
 pub mod util;
@@ -565,57 +561,4 @@ impl Client {
             }),
         }
     }
-
-    pub async fn ingest_aux_files(
-        &self,
-        tenant_shard_id: TenantShardId,
-        timeline_id: TimelineId,
-        aux_files: HashMap<String, String>,
-    ) -> Result<bool> {
-        let uri = format!(
-            "{}/v1/tenant/{}/timeline/{}/ingest_aux_files",
-            self.mgmt_api_endpoint, tenant_shard_id, timeline_id
-        );
-        let resp = self
-            .request_noerror(Method::POST, &uri, IngestAuxFilesRequest { aux_files })
-            .await?;
-        match resp.status() {
-            StatusCode::OK => Ok(true),
-            status => Err(match resp.json::<HttpErrorBody>().await {
-                Ok(HttpErrorBody { msg }) => Error::ApiError(status, msg),
-                Err(_) => {
-                    Error::ReceiveErrorBody(format!("Http error ({}) at {}.", status.as_u16(), uri))
-                }
-            }),
-        }
-    }
-
-    pub async fn list_aux_files(
-        &self,
-        tenant_shard_id: TenantShardId,
-        timeline_id: TimelineId,
-        lsn: Lsn,
-    ) -> Result<HashMap<String, Bytes>> {
-        let uri = format!(
-            "{}/v1/tenant/{}/timeline/{}/list_aux_files",
-            self.mgmt_api_endpoint, tenant_shard_id, timeline_id
-        );
-        let resp = self
-            .request_noerror(Method::POST, &uri, ListAuxFilesRequest { lsn })
-            .await?;
-        match resp.status() {
-            StatusCode::OK => {
-                let resp: HashMap<String, Bytes> = resp.json().await.map_err(|e| {
-                    Error::ApiError(StatusCode::INTERNAL_SERVER_ERROR, format!("{e}"))
-                })?;
-                Ok(resp)
-            }
-            status => Err(match resp.json::<HttpErrorBody>().await {
-                Ok(HttpErrorBody { msg }) => Error::ApiError(status, msg),
-                Err(_) => {
-                    Error::ReceiveErrorBody(format!("Http error ({}) at {}.", status.as_u16(), uri))
-                }
-            }),
-        }
-    }
 }

pageserver/compaction/src/bin/compaction-simulator.rs

@@ -1,5 +1,4 @@
 use clap::{Parser, Subcommand};
-use pageserver_compaction::helpers::PAGE_SZ;
 use pageserver_compaction::simulator::MockTimeline;
 use rand::Rng;
 use std::io::Write;
@@ -52,7 +51,7 @@ async fn simulate(cmd: &SimulateCmd, results_path: &Path) -> anyhow::Result<()>
     let mut executor = MockTimeline::new();
 
     // Convert the logical size in MB into a key range.
-    let key_range = 0..((cmd.logical_size * 1024 * 1024) / PAGE_SZ);
+    let key_range = 0..((cmd.logical_size * 1024 * 1024) / 8192);
     //let key_range = u64::MIN..u64::MAX;
     println!(
         "starting simulation with key range {:016X}-{:016X}",

pageserver/compaction/src/compact_tiered.rs

@@ -25,7 +25,7 @@ use std::collections::{HashSet, VecDeque};
 use std::ops::Range;
 
 use crate::helpers::{
-    accum_key_values, keyspace_total_size, merge_delta_keys_buffered, overlaps_with, PAGE_SZ,
+    accum_key_values, keyspace_total_size, merge_delta_keys_buffered, overlaps_with,
 };
 use crate::interface::*;
 use utils::lsn::Lsn;
@@ -379,7 +379,7 @@ where
                 .get_keyspace(&job.key_range, job.lsn_range.end, ctx)
                 .await?,
             &self.shard_identity,
-        ) * PAGE_SZ;
+        ) * 8192;
 
         let wal_size = job
             .input_layers
@@ -441,7 +441,7 @@ where
         let mut window = KeyspaceWindow::new(
             E::Key::MIN..E::Key::MAX,
             keyspace,
-            self.target_file_size / PAGE_SZ,
+            self.target_file_size / 8192,
         );
         while let Some(key_range) = window.choose_next_image(&self.shard_identity) {
             new_jobs.push(CompactionJob::<E> {
@@ -663,8 +663,8 @@ where
     }
 }
 
-/// Sliding window through keyspace and values for image layer
-/// This is used by [`LevelCompactionState::cover_with_images`] to decide on good split points
+// Sliding window through keyspace and values
+// This is used by over_with_images to decide on good split points
 struct KeyspaceWindow<K> {
     head: KeyspaceWindowHead<K>,
 
@@ -804,9 +804,9 @@ struct WindowElement<K> {
     accum_size: u64,
 }
 
-/// Sliding window through keyspace and values for delta layer tiling
-///
-/// This is used to decide which delta layer to write next.
+// Sliding window through keyspace and values
+//
+// This is used to decide what layer to write next, from the beginning of the window.
 struct Window<K> {
     elems: VecDeque<WindowElement<K>>,
 
@@ -830,13 +830,11 @@ where
     fn feed(&mut self, key: K, size: u64) {
         let last_size;
         if let Some(last) = self.elems.back_mut() {
-            // We require the keys to be strictly increasing for the window.
-            // Keys should already have been deduplicated by `accum_key_values`
-            assert!(
-                last.last_key < key,
-                "last_key(={}) >= key(={key})",
-                last.last_key
-            );
+            assert!(last.last_key <= key);
+            if key == last.last_key {
+                last.accum_size += size;
+                return;
+            }
             last_size = last.accum_size;
         } else {
             last_size = 0;
@@ -924,7 +922,7 @@ where
         // If we're willing to stretch it up to 1.25 target size, could we
         // gobble up the rest of the work? This avoids creating very small
         // "tail" layers at the end of the keyspace
-        if !has_more && self.remain_size() < target_size * 5 / 4 {
+        if !has_more && self.remain_size() < target_size * 5 / 3 {
             self.commit_upto(self.elems.len());
         } else {
             let delta_split_at = self.find_size_split(target_size);

pageserver/compaction/src/helpers.rs

@@ -16,8 +16,6 @@ use std::pin::Pin;
 use std::task::{ready, Poll};
 use utils::lsn::Lsn;
 
-pub const PAGE_SZ: u64 = 8192;
-
 pub fn keyspace_total_size<K>(
     keyspace: &CompactionKeySpace<K>,
     shard_identity: &ShardIdentity,

pageserver/compaction/src/simulator.rs

@@ -14,7 +14,6 @@ use std::ops::Range;
 use std::sync::Arc;
 use std::sync::Mutex;
 
-use crate::helpers::PAGE_SZ;
 use crate::helpers::{merge_delta_keys, overlaps_with};
 
 use crate::interface;
@@ -510,7 +509,7 @@ impl interface::CompactionJobExecutor for MockTimeline {
         let new_layer = Arc::new(MockImageLayer {
             key_range: key_range.clone(),
             lsn_range: lsn..lsn,
-            file_size: accum_size * PAGE_SZ,
+            file_size: accum_size * 8192,
             deleted: Mutex::new(false),
         });
         info!(

pageserver/ctl/src/main.rs

@@ -219,7 +219,6 @@ fn handle_metadata(
     let mut meta = TimelineMetadata::from_bytes(&metadata_bytes)?;
     println!("Current metadata:\n{meta:?}");
     let mut update_meta = false;
-    // TODO: simplify this part
     if let Some(disk_consistent_lsn) = disk_consistent_lsn {
         meta = TimelineMetadata::new(
             *disk_consistent_lsn,

pageserver/pagebench/src/cmd/aux_files.rs

@@ -1,98 +0,0 @@
-use pageserver_api::models::{AuxFilePolicy, TenantConfig, TenantConfigRequest};
-use pageserver_api::shard::TenantShardId;
-use utils::id::TenantTimelineId;
-use utils::lsn::Lsn;
-
-use std::collections::HashMap;
-use std::sync::Arc;
-
-/// Ingest aux files into the pageserver.
-#[derive(clap::Parser)]
-pub(crate) struct Args {
-    #[clap(long, default_value = "http://localhost:9898")]
-    mgmt_api_endpoint: String,
-    #[clap(long, default_value = "postgres://postgres@localhost:64000")]
-    page_service_connstring: String,
-    #[clap(long)]
-    pageserver_jwt: Option<String>,
-
-    targets: Option<Vec<TenantTimelineId>>,
-}
-
-pub(crate) fn main(args: Args) -> anyhow::Result<()> {
-    let rt = tokio::runtime::Builder::new_multi_thread()
-        .enable_all()
-        .build()
-        .unwrap();
-
-    let main_task = rt.spawn(main_impl(args));
-    rt.block_on(main_task).unwrap()
-}
-
-async fn main_impl(args: Args) -> anyhow::Result<()> {
-    let args: &'static Args = Box::leak(Box::new(args));
-
-    let mgmt_api_client = Arc::new(pageserver_client::mgmt_api::Client::new(
-        args.mgmt_api_endpoint.clone(),
-        args.pageserver_jwt.as_deref(),
-    ));
-
-    // discover targets
-    let timelines: Vec<TenantTimelineId> = crate::util::cli::targets::discover(
-        &mgmt_api_client,
-        crate::util::cli::targets::Spec {
-            limit_to_first_n_targets: None,
-            targets: {
-                if let Some(targets) = &args.targets {
-                    if targets.len() != 1 {
-                        anyhow::bail!("must specify exactly one target");
-                    }
-                    Some(targets.clone())
-                } else {
-                    None
-                }
-            },
-        },
-    )
-    .await?;
-
-    let timeline = timelines[0];
-    let tenant_shard_id = TenantShardId::unsharded(timeline.tenant_id);
-    let timeline_id = timeline.timeline_id;
-
-    println!("operating on timeline {}", timeline);
-
-    mgmt_api_client
-        .tenant_config(&TenantConfigRequest {
-            tenant_id: timeline.tenant_id,
-            config: TenantConfig {
-                switch_aux_file_policy: Some(AuxFilePolicy::V2),
-                ..Default::default()
-            },
-        })
-        .await?;
-
-    for batch in 0..100 {
-        let items = (0..100)
-            .map(|id| {
-                (
-                    format!("pg_logical/mappings/{:03}.{:03}", batch, id),
-                    format!("{:08}", id),
-                )
-            })
-            .collect::<HashMap<_, _>>();
-        let file_cnt = items.len();
-        mgmt_api_client
-            .ingest_aux_files(tenant_shard_id, timeline_id, items)
-            .await?;
-        println!("ingested {file_cnt} files");
-    }
-
-    let files = mgmt_api_client
-        .list_aux_files(tenant_shard_id, timeline_id, Lsn(Lsn::MAX.0 - 1))
-        .await?;
-
-    println!("{} files found", files.len());
-
-    anyhow::Ok(())
-}

pageserver/pagebench/src/main.rs

@@ -14,7 +14,6 @@ mod util {
 
 /// The pagebench CLI sub-commands, dispatched in [`main`] below.
 mod cmd {
-    pub(super) mod aux_files;
     pub(super) mod basebackup;
     pub(super) mod getpage_latest_lsn;
     pub(super) mod ondemand_download_churn;
@@ -28,7 +27,6 @@ enum Args {
     GetPageLatestLsn(cmd::getpage_latest_lsn::Args),
     TriggerInitialSizeCalculation(cmd::trigger_initial_size_calculation::Args),
     OndemandDownloadChurn(cmd::ondemand_download_churn::Args),
-    AuxFiles(cmd::aux_files::Args),
 }
 
 fn main() {
@@ -48,7 +46,6 @@ fn main() {
             cmd::trigger_initial_size_calculation::main(args)
         }
         Args::OndemandDownloadChurn(args) => cmd::ondemand_download_churn::main(args),
-        Args::AuxFiles(args) => cmd::aux_files::main(args),
     }
     .unwrap()
 }

pageserver/src/disk_usage_eviction_task.rs

@@ -535,11 +535,17 @@ pub(crate) async fn disk_usage_eviction_task_iteration_impl<U: Usage>(
                 }
                 EvictionLayer::Secondary(layer) => {
                     let file_size = layer.metadata.file_size();
+                    let tenant_manager = tenant_manager.clone();
 
                     js.spawn(async move {
                         layer
                             .secondary_tenant
-                            .evict_layer(layer.timeline_id, layer.name)
+                            .evict_layer(
+                                tenant_manager.get_conf(),
+                                layer.timeline_id,
+                                layer.name,
+                                layer.metadata,
+                            )
                             .await;
                         Ok(file_size)
                     });

pageserver/src/http/openapi_spec.yml

@@ -257,37 +257,6 @@ paths:
               schema:
                 $ref: "#/components/schemas/LsnByTimestampResponse"
 
-  /v1/tenant/{tenant_shard_id}/timeline/{timeline_id}/lsn_lease:
-    parameters:
-      - name: tenant_shard_id
-        in: path
-        required: true
-        schema:
-          type: string
-      - name: timeline_id
-        in: path
-        required: true
-        schema:
-          type: string
-          format: hex
-    post:
-      description: Obtain lease for the given LSN
-      parameters:
-        - name: lsn
-          in: query
-          required: true
-          schema:
-            type: string
-            format: hex
-          description: A LSN to obtain the lease for
-      responses:
-        "200":
-          description: OK
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/LsnLease"
-
   /v1/tenant/{tenant_id}/timeline/{timeline_id}/do_gc:
     parameters:
       - name: tenant_id
@@ -612,80 +581,6 @@ paths:
               schema:
                 $ref: "#/components/schemas/Error"
 
-  /v1/tenant/{tenant_shard_id}/timeline/{timeline_id}/detach_ancestor:
-    parameters:
-      - name: tenant_shard_id
-        in: path
-        required: true
-        schema:
-          type: string
-      - name: timeline_id
-        in: path
-        ŕequired: true
-        schema:
-          type: string
-
-    put:
-      description: |
-        Detach a timeline from its ancestor and reparent all ancestors timelines with lower `ancestor_lsn`.
-        Current implementation might not be retryable across failure cases, but will be enhanced in future.
-        Detaching should be expected to be expensive operation. Timeouts should be retried.
-      responses:
-        "200":
-          description: |
-            The timeline has been detached from it's ancestor (now or earlier), and at least the returned timelines have been reparented.
-            If any timelines were deleted after reparenting, they might not be on this list.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/AncestorDetached"
-
-        "400":
-          description: |
-            Number of early checks meaning the timeline cannot be detached now:
-              - the ancestor of timeline has an ancestor: not supported, see RFC
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/Error"
-
-        "404":
-          description: Tenant or timeline not found.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/NotFoundError"
-
-        "409":
-          description: |
-            The timeline can never be detached:
-              - timeline has no ancestor, implying that the timeline has never had an ancestor
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ConflictError"
-
-        "500":
-          description: |
-            Transient error, for example, pageserver shutdown happened while
-            processing the request but we were unable to distinguish that. Must
-            be retried.
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/Error"
-
-        "503":
-          description: |
-            Temporarily unavailable, please retry. Possible reasons:
-              - another timeline detach for the same tenant is underway, please retry later
-              - detected shutdown error
-          content:
-            application/json:
-              schema:
-                $ref: "#/components/schemas/ServiceUnavailableError"
-
-
   /v1/tenant/:
     get:
       description: Get tenants list
@@ -1085,15 +980,6 @@ components:
           type: string
           enum: [past, present, future, nodata]
 
-    LsnLease:
-      type: object
-      required:
-        - valid_until
-      properties:
-        valid_until:
-          type: string
-          format: date-time
-
     PageserverUtilization:
       type: object
       required:
@@ -1151,19 +1037,6 @@ components:
           format: int64
           description: How many bytes of layer content were in the latest layer heatmap
 
-    AncestorDetached:
-      type: object
-      required:
-        - reparented_timelines
-      properties:
-        reparented_timelines:
-          type: array
-          description: Set of reparented timeline ids
-          properties:
-            type: string
-            format: hex
-            description: TimelineId
-
 
     Error:
       type: object

pageserver/src/http/routes.rs

@@ -16,9 +16,6 @@ use hyper::header;
 use hyper::StatusCode;
 use hyper::{Body, Request, Response, Uri};
 use metrics::launch_timestamp::LaunchTimestamp;
-use pageserver_api::models::AuxFilePolicy;
-use pageserver_api::models::IngestAuxFilesRequest;
-use pageserver_api::models::ListAuxFilesRequest;
 use pageserver_api::models::LocationConfig;
 use pageserver_api::models::LocationConfigListResponse;
 use pageserver_api::models::ShardParameters;
@@ -75,7 +72,6 @@ use crate::tenant::storage_layer::LayerAccessStatsReset;
 use crate::tenant::storage_layer::LayerName;
 use crate::tenant::timeline::CompactFlags;
 use crate::tenant::timeline::Timeline;
-use crate::tenant::GetTimelineError;
 use crate::tenant::SpawnMode;
 use crate::tenant::{LogicalSizeCalculationCause, PageReconstructError};
 use crate::{config::PageServerConf, tenant::mgr};
@@ -281,13 +277,6 @@ impl From<GetTenantError> for ApiError {
     }
 }
 
-impl From<GetTimelineError> for ApiError {
-    fn from(gte: GetTimelineError) -> Self {
-        // Rationale: tenant is activated only after eligble timelines activate
-        ApiError::NotFound(gte.into())
-    }
-}
-
 impl From<GetActiveTenantError> for ApiError {
     fn from(e: GetActiveTenantError) -> ApiError {
         match e {
@@ -395,7 +384,7 @@ async fn build_timeline_info_common(
         let guard = timeline.last_received_wal.lock().unwrap();
         if let Some(info) = guard.as_ref() {
             (
-                Some(format!("{}", info.wal_source_connconf)), // Password is hidden, but it's for statistics only.
+                Some(format!("{:?}", info.wal_source_connconf)), // Password is hidden, but it's for statistics only.
                 Some(info.last_received_msg_lsn),
                 Some(info.last_received_msg_ts),
             )
@@ -450,8 +439,6 @@ async fn build_timeline_info_common(
         state,
 
         walreceiver_status,
-
-        last_aux_file_policy: timeline.last_aux_file_policy.load(),
     };
     Ok(info)
 }
@@ -652,7 +639,9 @@ async fn timeline_preserve_initdb_handler(
             .tenant_manager
             .get_attached_tenant_shard(tenant_shard_id)?;
 
-        let timeline = tenant.get_timeline(timeline_id, false)?;
+        let timeline = tenant
+            .get_timeline(timeline_id, false)
+            .map_err(|e| ApiError::NotFound(e.into()))?;
 
         timeline
             .preserve_initdb_archive()
@@ -694,7 +683,9 @@ async fn timeline_detail_handler(
 
         tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
 
-        let timeline = tenant.get_timeline(timeline_id, false)?;
+        let timeline = tenant
+            .get_timeline(timeline_id, false)
+            .map_err(|e| ApiError::NotFound(e.into()))?;
 
         let timeline_info = build_timeline_info(
             &timeline,
@@ -1706,32 +1697,6 @@ async fn handle_tenant_break(
     json_response(StatusCode::OK, ())
 }
 
-// Obtains an lsn lease on the given timeline.
-async fn lsn_lease_handler(
-    request: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
-    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
-    check_permission(&request, Some(tenant_shard_id.tenant_id))?;
-
-    let lsn: Lsn = parse_query_param(&request, "lsn")?
-        .ok_or_else(|| ApiError::BadRequest(anyhow!("missing 'lsn' query parameter")))?;
-
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
-
-    let state = get_state(&request);
-
-    let timeline =
-        active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
-            .await?;
-    let result = timeline
-        .make_lsn_lease(lsn, &ctx)
-        .map_err(|e| ApiError::InternalServerError(e.context("lsn lease http handler")))?;
-
-    json_response(StatusCode::OK, result)
-}
-
 // Run GC immediately on given timeline.
 async fn timeline_gc_handler(
     mut request: Request<Body>,
@@ -1767,8 +1732,6 @@ async fn timeline_compact_handler(
     if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
         flags |= CompactFlags::ForceImageLayerCreation;
     }
-    let wait_until_uploaded =
-        parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
 
     async {
         let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
@@ -1777,9 +1740,6 @@ async fn timeline_compact_handler(
             .compact(&cancel, flags, &ctx)
             .await
             .map_err(|e| ApiError::InternalServerError(e.into()))?;
-        if wait_until_uploaded {
-            timeline.remote_client.wait_completion().await.map_err(ApiError::InternalServerError)?;
-        }
         json_response(StatusCode::OK, ())
     }
     .instrument(info_span!("manual_compaction", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
@@ -1804,8 +1764,6 @@ async fn timeline_checkpoint_handler(
     if Some(true) == parse_query_param::<_, bool>(&request, "force_image_layer_creation")? {
         flags |= CompactFlags::ForceImageLayerCreation;
     }
-    let wait_until_uploaded =
-        parse_query_param::<_, bool>(&request, "wait_until_uploaded")?.unwrap_or(false);
 
     async {
         let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
@@ -1819,10 +1777,6 @@ async fn timeline_checkpoint_handler(
             .await
             .map_err(|e| ApiError::InternalServerError(e.into()))?;
 
-        if wait_until_uploaded {
-            timeline.remote_client.wait_completion().await.map_err(ApiError::InternalServerError)?;
-        }
-
         json_response(StatusCode::OK, ())
     }
     .instrument(info_span!("manual_checkpoint", tenant_id = %tenant_shard_id.tenant_id, shard_id = %tenant_shard_id.shard_slug(), %timeline_id))
@@ -1906,11 +1860,14 @@ async fn timeline_detach_ancestor_handler(
         let ctx = RequestContext::new(TaskKind::DetachAncestor, DownloadBehavior::Download);
         let ctx = &ctx;
 
-        let timeline = tenant.get_timeline(timeline_id, true)?;
+        let timeline = tenant
+            .get_timeline(timeline_id, true)
+            .map_err(|e| ApiError::NotFound(e.into()))?;
 
         let (_guard, prepared) = timeline
             .prepare_to_detach_from_ancestor(&tenant, options, ctx)
-            .await?;
+            .await
+            .map_err(|e| ApiError::InternalServerError(e.into()))?;
 
         let res = state
             .tenant_manager
@@ -2044,7 +2001,9 @@ async fn active_timeline_of_active_tenant(
 
     tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
 
-    Ok(tenant.get_timeline(timeline_id, true)?)
+    tenant
+        .get_timeline(timeline_id, true)
+        .map_err(|e| ApiError::NotFound(e.into()))
 }
 
 async fn always_panic_handler(
@@ -2308,31 +2267,6 @@ async fn post_tracing_event_handler(
     json_response(StatusCode::OK, ())
 }
 
-async fn force_aux_policy_switch_handler(
-    mut r: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    check_permission(&r, None)?;
-    let tenant_shard_id: TenantShardId = parse_request_param(&r, "tenant_shard_id")?;
-    let timeline_id: TimelineId = parse_request_param(&r, "timeline_id")?;
-    let policy: AuxFilePolicy = json_request(&mut r).await?;
-
-    let state = get_state(&r);
-
-    let tenant = state
-        .tenant_manager
-        .get_attached_tenant_shard(tenant_shard_id)?;
-    tenant.wait_to_become_active(ACTIVE_TENANT_TIMEOUT).await?;
-    let timeline =
-        active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
-            .await?;
-    timeline
-        .do_switch_aux_policy(policy)
-        .map_err(ApiError::InternalServerError)?;
-
-    json_response(StatusCode::OK, ())
-}
-
 async fn put_io_engine_handler(
     mut r: Request<Body>,
     _cancel: CancellationToken,
@@ -2395,59 +2329,6 @@ async fn get_utilization(
         .map_err(ApiError::InternalServerError)
 }
 
-async fn list_aux_files(
-    mut request: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
-    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
-    let body: ListAuxFilesRequest = json_request(&mut request).await?;
-    check_permission(&request, Some(tenant_shard_id.tenant_id))?;
-
-    let state = get_state(&request);
-
-    let timeline =
-        active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
-            .await?;
-
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
-    let files = timeline.list_aux_files(body.lsn, &ctx).await?;
-    json_response(StatusCode::OK, files)
-}
-
-async fn ingest_aux_files(
-    mut request: Request<Body>,
-    _cancel: CancellationToken,
-) -> Result<Response<Body>, ApiError> {
-    let tenant_shard_id: TenantShardId = parse_request_param(&request, "tenant_shard_id")?;
-    let timeline_id: TimelineId = parse_request_param(&request, "timeline_id")?;
-    let body: IngestAuxFilesRequest = json_request(&mut request).await?;
-    check_permission(&request, Some(tenant_shard_id.tenant_id))?;
-
-    let state = get_state(&request);
-
-    let timeline =
-        active_timeline_of_active_tenant(&state.tenant_manager, tenant_shard_id, timeline_id)
-            .await?;
-
-    let mut modification = timeline.begin_modification(
-        Lsn(timeline.get_last_record_lsn().0 + 8), /* advance LSN by 8 */
-    );
-    let ctx = RequestContext::new(TaskKind::MgmtRequest, DownloadBehavior::Download);
-    for (fname, content) in body.aux_files {
-        modification
-            .put_file(&fname, content.as_bytes(), &ctx)
-            .await
-            .map_err(ApiError::InternalServerError)?;
-    }
-    modification
-        .commit(&ctx)
-        .await
-        .map_err(ApiError::InternalServerError)?;
-
-    json_response(StatusCode::OK, ())
-}
-
 /// Report on the largest tenants on this pageserver, for the storage controller to identify
 /// candidates for splitting
 async fn post_top_tenants(
@@ -2751,10 +2632,6 @@ pub fn make_router(
             "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/get_timestamp_of_lsn",
             |r| api_handler(r, get_timestamp_of_lsn_handler),
         )
-        .post(
-            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/lsn_lease",
-            |r| api_handler(r, lsn_lease_handler),
-        )
         .put(
             "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/do_gc",
             |r| api_handler(r, timeline_gc_handler),
@@ -2828,19 +2705,7 @@ pub fn make_router(
             |r| api_handler(r, timeline_collect_keyspace),
         )
         .put("/v1/io_engine", |r| api_handler(r, put_io_engine_handler))
-        .put(
-            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/force_aux_policy_switch",
-            |r| api_handler(r, force_aux_policy_switch_handler),
-        )
         .get("/v1/utilization", |r| api_handler(r, get_utilization))
-        .post(
-            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/ingest_aux_files",
-            |r| testing_api_handler("ingest_aux_files", r, ingest_aux_files),
-        )
-        .post(
-            "/v1/tenant/:tenant_shard_id/timeline/:timeline_id/list_aux_files",
-            |r| testing_api_handler("list_aux_files", r, list_aux_files),
-        )
         .post("/v1/top_tenants", |r| api_handler(r, post_top_tenants))
         .any(handler_404))
 }

pageserver/src/metrics.rs

@@ -525,15 +525,6 @@ static LAST_RECORD_LSN: Lazy<IntGaugeVec> = Lazy::new(|| {
     .expect("failed to define a metric")
 });
 
-static STANDBY_HORIZON: Lazy<IntGaugeVec> = Lazy::new(|| {
-    register_int_gauge_vec!(
-        "pageserver_standby_horizon",
-        "Standby apply LSN for which GC is hold off, by timeline.",
-        &["tenant_id", "shard_id", "timeline_id"]
-    )
-    .expect("failed to define a metric")
-});
-
 static RESIDENT_PHYSICAL_SIZE: Lazy<UIntGaugeVec> = Lazy::new(|| {
     register_uint_gauge_vec!(
         "pageserver_resident_physical_size",
@@ -1867,6 +1858,7 @@ pub(crate) struct WalIngestMetrics {
     pub(crate) records_received: IntCounter,
     pub(crate) records_committed: IntCounter,
     pub(crate) records_filtered: IntCounter,
+    pub(crate) time_spent_on_ingest: Histogram,
 }
 
 pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMetrics {
@@ -1890,6 +1882,12 @@ pub(crate) static WAL_INGEST: Lazy<WalIngestMetrics> = Lazy::new(|| WalIngestMet
         "Number of WAL records filtered out due to sharding"
     )
     .expect("failed to define a metric"),
+    time_spent_on_ingest: register_histogram!(
+        "pageserver_wal_ingest_put_value_seconds",
+        "Actual time spent on ingesting a record",
+        redo_histogram_time_buckets!(),
+    )
+    .expect("failed to define a metric"),
 });
 
 pub(crate) static WAL_REDO_TIME: Lazy<Histogram> = Lazy::new(|| {
@@ -2100,7 +2098,6 @@ pub(crate) struct TimelineMetrics {
     pub garbage_collect_histo: StorageTimeMetrics,
     pub find_gc_cutoffs_histo: StorageTimeMetrics,
     pub last_record_gauge: IntGauge,
-    pub standby_horizon_gauge: IntGauge,
     pub resident_physical_size_gauge: UIntGauge,
     /// copy of LayeredTimeline.current_logical_size
     pub current_logical_size_gauge: UIntGauge,
@@ -2170,9 +2167,6 @@ impl TimelineMetrics {
         let last_record_gauge = LAST_RECORD_LSN
             .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
             .unwrap();
-        let standby_horizon_gauge = STANDBY_HORIZON
-            .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
-            .unwrap();
         let resident_physical_size_gauge = RESIDENT_PHYSICAL_SIZE
             .get_metric_with_label_values(&[&tenant_id, &shard_id, &timeline_id])
             .unwrap();
@@ -2218,7 +2212,6 @@ impl TimelineMetrics {
             find_gc_cutoffs_histo,
             load_layer_map_histo,
             last_record_gauge,
-            standby_horizon_gauge,
             resident_physical_size_gauge,
             current_logical_size_gauge,
             aux_file_size_gauge,
@@ -2253,7 +2246,6 @@ impl TimelineMetrics {
         let timeline_id = &self.timeline_id;
         let shard_id = &self.shard_id;
         let _ = LAST_RECORD_LSN.remove_label_values(&[tenant_id, shard_id, timeline_id]);
-        let _ = STANDBY_HORIZON.remove_label_values(&[tenant_id, shard_id, timeline_id]);
         {
             RESIDENT_PHYSICAL_SIZE_GLOBAL.sub(self.resident_physical_size_get());
             let _ = RESIDENT_PHYSICAL_SIZE.remove_label_values(&[tenant_id, shard_id, timeline_id]);

pageserver/src/page_service.rs

@@ -19,7 +19,6 @@ use pageserver_api::models::{
 };
 use pageserver_api::shard::ShardIndex;
 use pageserver_api::shard::ShardNumber;
-use pageserver_api::shard::TenantShardId;
 use postgres_backend::{is_expected_io_error, AuthType, PostgresBackend, QueryError};
 use pq_proto::framed::ConnectionError;
 use pq_proto::FeStartupPacket;
@@ -34,7 +33,6 @@ use std::str::FromStr;
 use std::sync::Arc;
 use std::time::Duration;
 use std::time::Instant;
-use std::time::SystemTime;
 use tokio::io::AsyncWriteExt;
 use tokio::io::{AsyncRead, AsyncWrite};
 use tokio_util::io::StreamReader;
@@ -907,39 +905,6 @@ impl PageServerHandler {
         }
     }
 
-    #[instrument(skip_all, fields(shard_id, %lsn))]
-    async fn handle_make_lsn_lease<IO>(
-        &self,
-        pgb: &mut PostgresBackend<IO>,
-        tenant_shard_id: TenantShardId,
-        timeline_id: TimelineId,
-        lsn: Lsn,
-        ctx: &RequestContext,
-    ) -> Result<(), QueryError>
-    where
-        IO: AsyncRead + AsyncWrite + Send + Sync + Unpin,
-    {
-        let shard_selector = ShardSelector::Known(tenant_shard_id.to_index());
-        let timeline = self
-            .get_active_tenant_timeline(tenant_shard_id.tenant_id, timeline_id, shard_selector)
-            .await?;
-        let lease = timeline.make_lsn_lease(lsn, ctx)?;
-        let valid_until = lease
-            .valid_until
-            .duration_since(SystemTime::UNIX_EPOCH)
-            .map_err(|e| QueryError::Other(e.into()))?;
-
-        pgb.write_message_noflush(&BeMessage::RowDescription(&[RowDescriptor::text_col(
-            b"valid_until",
-        )]))?
-        .write_message_noflush(&BeMessage::DataRow(&[Some(
-            &valid_until.as_millis().to_be_bytes(),
-        )]))?
-        .write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
-
-        Ok(())
-    }
-
     #[instrument(skip_all, fields(shard_id))]
     async fn handle_get_rel_exists_request(
         &mut self,
@@ -1521,8 +1486,9 @@ where
 
         let ctx = self.connection_ctx.attached_child();
         debug!("process query {query_string:?}");
-        let parts = query_string.split_whitespace().collect::<Vec<_>>();
-        if let Some(params) = parts.strip_prefix(&["pagestream_v2"]) {
+        if query_string.starts_with("pagestream_v2 ") {
+            let (_, params_raw) = query_string.split_at("pagestream_v2 ".len());
+            let params = params_raw.split(' ').collect::<Vec<_>>();
             if params.len() != 2 {
                 return Err(QueryError::Other(anyhow::anyhow!(
                     "invalid param number for pagestream command"
@@ -1547,7 +1513,9 @@ where
                 ctx,
             )
             .await?;
-        } else if let Some(params) = parts.strip_prefix(&["pagestream"]) {
+        } else if query_string.starts_with("pagestream ") {
+            let (_, params_raw) = query_string.split_at("pagestream ".len());
+            let params = params_raw.split(' ').collect::<Vec<_>>();
             if params.len() != 2 {
                 return Err(QueryError::Other(anyhow::anyhow!(
                     "invalid param number for pagestream command"
@@ -1572,7 +1540,10 @@ where
                 ctx,
             )
             .await?;
-        } else if let Some(params) = parts.strip_prefix(&["basebackup"]) {
+        } else if query_string.starts_with("basebackup ") {
+            let (_, params_raw) = query_string.split_at("basebackup ".len());
+            let params = params_raw.split_whitespace().collect::<Vec<_>>();
+
             if params.len() < 2 {
                 return Err(QueryError::Other(anyhow::anyhow!(
                     "invalid param number for basebackup command"
@@ -1590,23 +1561,26 @@ where
 
             self.check_permission(Some(tenant_id))?;
 
-            let lsn = if let Some(lsn_str) = params.get(2) {
+            let lsn = if params.len() >= 3 {
                 Some(
-                    Lsn::from_str(lsn_str)
-                        .with_context(|| format!("Failed to parse Lsn from {lsn_str}"))?,
+                    Lsn::from_str(params[2])
+                        .with_context(|| format!("Failed to parse Lsn from {}", params[2]))?,
                 )
             } else {
                 None
             };
 
-            let gzip = match params.get(3) {
-                Some(&"--gzip") => true,
-                None => false,
-                Some(third_param) => {
+            let gzip = if params.len() >= 4 {
+                if params[3] == "--gzip" {
+                    true
+                } else {
                     return Err(QueryError::Other(anyhow::anyhow!(
-                        "Parameter in position 3 unknown {third_param}",
-                    )))
+                        "Parameter in position 3 unknown {}",
+                        params[3],
+                    )));
                 }
+            } else {
+                false
             };
 
             let metric_recording = metrics::BASEBACKUP_QUERY_TIME.start_recording(&ctx);
@@ -1630,7 +1604,10 @@ where
             res?;
         }
         // return pair of prev_lsn and last_lsn
-        else if let Some(params) = parts.strip_prefix(&["get_last_record_rlsn"]) {
+        else if query_string.starts_with("get_last_record_rlsn ") {
+            let (_, params_raw) = query_string.split_at("get_last_record_rlsn ".len());
+            let params = params_raw.split_whitespace().collect::<Vec<_>>();
+
             if params.len() != 2 {
                 return Err(QueryError::Other(anyhow::anyhow!(
                     "invalid param number for get_last_record_rlsn command"
@@ -1672,7 +1649,10 @@ where
             .await?;
         }
         // same as basebackup, but result includes relational data as well
-        else if let Some(params) = parts.strip_prefix(&["fullbackup"]) {
+        else if query_string.starts_with("fullbackup ") {
+            let (_, params_raw) = query_string.split_at("fullbackup ".len());
+            let params = params_raw.split_whitespace().collect::<Vec<_>>();
+
             if params.len() < 2 {
                 return Err(QueryError::Other(anyhow::anyhow!(
                     "invalid param number for fullbackup command"
@@ -1689,18 +1669,18 @@ where
                 .record("timeline_id", field::display(timeline_id));
 
             // The caller is responsible for providing correct lsn and prev_lsn.
-            let lsn = if let Some(lsn_str) = params.get(2) {
+            let lsn = if params.len() > 2 {
                 Some(
-                    Lsn::from_str(lsn_str)
-                        .with_context(|| format!("Failed to parse Lsn from {lsn_str}"))?,
+                    Lsn::from_str(params[2])
+                        .with_context(|| format!("Failed to parse Lsn from {}", params[2]))?,
                 )
             } else {
                 None
             };
-            let prev_lsn = if let Some(prev_lsn_str) = params.get(3) {
+            let prev_lsn = if params.len() > 3 {
                 Some(
-                    Lsn::from_str(prev_lsn_str)
-                        .with_context(|| format!("Failed to parse Lsn from {prev_lsn_str}"))?,
+                    Lsn::from_str(params[3])
+                        .with_context(|| format!("Failed to parse Lsn from {}", params[3]))?,
                 )
             } else {
                 None
@@ -1733,7 +1713,8 @@ where
             // 2. Run:
             // cat my_backup/base.tar | psql -h $PAGESERVER \
             //     -c "import basebackup $TENANT $TIMELINE $START_LSN $END_LSN $PG_VERSION"
-            let params = &parts[2..];
+            let (_, params_raw) = query_string.split_at("import basebackup ".len());
+            let params = params_raw.split_whitespace().collect::<Vec<_>>();
             if params.len() != 5 {
                 return Err(QueryError::Other(anyhow::anyhow!(
                     "invalid param number for import basebackup command"
@@ -1782,7 +1763,8 @@ where
             //
             // Files are scheduled to be persisted to remote storage, and the
             // caller should poll the http api to check when that is done.
-            let params = &parts[2..];
+            let (_, params_raw) = query_string.split_at("import wal ".len());
+            let params = params_raw.split_whitespace().collect::<Vec<_>>();
             if params.len() != 4 {
                 return Err(QueryError::Other(anyhow::anyhow!(
                     "invalid param number for import wal command"
@@ -1820,45 +1802,10 @@ where
             // important because psycopg2 executes "SET datestyle TO 'ISO'"
             // on connect
             pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?;
-        } else if query_string.starts_with("lease lsn ") {
-            let params = &parts[2..];
-            if params.len() != 3 {
-                return Err(QueryError::Other(anyhow::anyhow!(
-                    "invalid param number {} for lease lsn command",
-                    params.len()
-                )));
-            }
-
-            let tenant_shard_id = TenantShardId::from_str(params[0])
-                .with_context(|| format!("Failed to parse tenant id from {}", params[0]))?;
-            let timeline_id = TimelineId::from_str(params[1])
-                .with_context(|| format!("Failed to parse timeline id from {}", params[1]))?;
-
-            tracing::Span::current()
-                .record("tenant_id", field::display(tenant_shard_id))
-                .record("timeline_id", field::display(timeline_id));
-
-            self.check_permission(Some(tenant_shard_id.tenant_id))?;
-
-            // The caller is responsible for providing correct lsn.
-            let lsn = Lsn::from_str(params[2])
-                .with_context(|| format!("Failed to parse Lsn from {}", params[2]))?;
-
-            match self
-                .handle_make_lsn_lease(pgb, tenant_shard_id, timeline_id, lsn, &ctx)
-                .await
-            {
-                Ok(()) => pgb.write_message_noflush(&BeMessage::CommandComplete(b"SELECT 1"))?,
-                Err(e) => {
-                    error!("error obtaining lsn lease for {lsn}: {e:?}");
-                    pgb.write_message_noflush(&BeMessage::ErrorResponse(
-                        &e.to_string(),
-                        Some(e.pg_error_code()),
-                    ))?
-                }
-            };
-        } else if let Some(params) = parts.strip_prefix(&["show"]) {
+        } else if query_string.starts_with("show ") {
             // show <tenant_id>
+            let (_, params_raw) = query_string.split_at("show ".len());
+            let params = params_raw.split(' ').collect::<Vec<_>>();
             if params.len() != 1 {
                 return Err(QueryError::Other(anyhow::anyhow!(
                     "invalid param number for config command"

pageserver/src/pgdatadir_mapping.rs

@@ -9,6 +9,7 @@
 use super::tenant::{PageReconstructError, Timeline};
 use crate::context::RequestContext;
 use crate::keyspace::{KeySpace, KeySpaceAccum};
+use crate::metrics::WAL_INGEST;
 use crate::span::debug_assert_current_span_has_tenant_and_timeline_id_no_shard_id;
 use crate::walrecord::NeonWalRecord;
 use crate::{aux_file, repository::*};
@@ -34,16 +35,12 @@ use std::ops::ControlFlow;
 use std::ops::Range;
 use strum::IntoEnumIterator;
 use tokio_util::sync::CancellationToken;
-use tracing::{debug, info, trace, warn};
+use tracing::{debug, trace, warn};
 use utils::bin_ser::DeserializeError;
 use utils::vec_map::{VecMap, VecMapOrdering};
 use utils::{bin_ser::BeSer, lsn::Lsn};
 
-/// Max delta records appended to the AUX_FILES_KEY (for aux v1). The write path will write a full image once this threshold is reached.
-pub const MAX_AUX_FILE_DELTAS: usize = 1024;
-
-/// Max number of aux-file-related delta layers. The compaction will create a new image layer once this threshold is reached.
-pub const MAX_AUX_FILE_V2_DELTAS: usize = 64;
+const MAX_AUX_FILE_DELTAS: usize = 1024;
 
 #[derive(Debug)]
 pub enum LsnForTimestamp {
@@ -721,11 +718,10 @@ impl Timeline {
         lsn: Lsn,
         ctx: &RequestContext,
     ) -> Result<HashMap<String, Bytes>, PageReconstructError> {
-        let current_policy = self.last_aux_file_policy.load();
-        match current_policy {
-            Some(AuxFilePolicy::V1) | None => self.list_aux_files_v1(lsn, ctx).await,
-            Some(AuxFilePolicy::V2) => self.list_aux_files_v2(lsn, ctx).await,
-            Some(AuxFilePolicy::CrossValidation) => {
+        match self.get_switch_aux_file_policy() {
+            AuxFilePolicy::V1 => self.list_aux_files_v1(lsn, ctx).await,
+            AuxFilePolicy::V2 => self.list_aux_files_v2(lsn, ctx).await,
+            AuxFilePolicy::CrossValidation => {
                 let v1_result = self.list_aux_files_v1(lsn, ctx).await;
                 let v2_result = self.list_aux_files_v2(lsn, ctx).await;
                 match (v1_result, v2_result) {
@@ -1473,40 +1469,7 @@ impl<'a> DatadirModification<'a> {
         content: &[u8],
         ctx: &RequestContext,
     ) -> anyhow::Result<()> {
-        let switch_policy = self.tline.get_switch_aux_file_policy();
-
-        let policy = {
-            let current_policy = self.tline.last_aux_file_policy.load();
-            // Allowed switch path:
-            // * no aux files -> v1/v2/cross-validation
-            // * cross-validation->v2
-
-            let current_policy = if current_policy.is_none() {
-                // This path will only be hit once per tenant: we will decide the final policy in this code block.
-                // The next call to `put_file` will always have `last_aux_file_policy != None`.
-                let lsn = Lsn::max(self.tline.get_last_record_lsn(), self.lsn);
-                let aux_files_key_v1 = self.tline.list_aux_files_v1(lsn, ctx).await?;
-                if aux_files_key_v1.is_empty() {
-                    None
-                } else {
-                    self.tline.do_switch_aux_policy(AuxFilePolicy::V1)?;
-                    Some(AuxFilePolicy::V1)
-                }
-            } else {
-                current_policy
-            };
-
-            if AuxFilePolicy::is_valid_migration_path(current_policy, switch_policy) {
-                self.tline.do_switch_aux_policy(switch_policy)?;
-                info!(current=?current_policy, next=?switch_policy, "switching aux file policy");
-                switch_policy
-            } else {
-                // This branch handles non-valid migration path, and the case that switch_policy == current_policy.
-                // And actually, because the migration path always allow unspecified -> *, this unwrap_or will never be hit.
-                current_policy.unwrap_or(AuxFilePolicy::default_tenant_config())
-            }
-        };
-
+        let policy = self.tline.get_switch_aux_file_policy();
         if let AuxFilePolicy::V2 | AuxFilePolicy::CrossValidation = policy {
             let key = aux_file::encode_aux_file_key(path);
             // retrieve the key from the engine
@@ -1714,6 +1677,8 @@ impl<'a> DatadirModification<'a> {
     pub async fn commit(&mut self, ctx: &RequestContext) -> anyhow::Result<()> {
         let mut writer = self.tline.writer().await;
 
+        let timer = WAL_INGEST.time_spent_on_ingest.start_timer();
+
         let pending_nblocks = self.pending_nblocks;
         self.pending_nblocks = 0;
 
@@ -1753,6 +1718,8 @@ impl<'a> DatadirModification<'a> {
             writer.update_directory_entries_count(kind, count as u64);
         }
 
+        timer.observe_duration();
+
         Ok(())
     }
 
@@ -1788,12 +1755,6 @@ impl<'a> DatadirModification<'a> {
         self.tline.get(key, lsn, ctx).await
     }
 
-    /// Only used during unit tests, force putting a key into the modification.
-    #[cfg(test)]
-    pub(crate) fn put_for_test(&mut self, key: Key, val: Value) {
-        self.put(key, val);
-    }
-
     fn put(&mut self, key: Key, val: Value) {
         let values = self.pending_updates.entry(key).or_default();
         // Replace the previous value if it exists at the same lsn

pageserver/src/tenant/blob_io.rs

@@ -238,13 +238,10 @@ impl<const BUFFERED: bool> BlobWriter<BUFFERED> {
                         io_buf,
                         Err(Error::new(
                             ErrorKind::Other,
-                            format!("blob too large ({len} bytes)"),
+                            format!("blob too large ({} bytes)", len),
                         )),
                     );
                 }
-                if len > 0x0fff_ffff {
-                    tracing::warn!("writing blob above future limit ({len} bytes)");
-                }
                 let mut len_buf = (len as u32).to_be_bytes();
                 len_buf[0] |= 0x80;
                 io_buf.extend_from_slice(&len_buf[..]);

pageserver/src/tenant/config.rs

@@ -11,7 +11,6 @@
 use anyhow::bail;
 use pageserver_api::models::AuxFilePolicy;
 use pageserver_api::models::CompactionAlgorithm;
-use pageserver_api::models::CompactionAlgorithmSettings;
 use pageserver_api::models::EvictionPolicy;
 use pageserver_api::models::{self, ThrottleConfig};
 use pageserver_api::shard::{ShardCount, ShardIdentity, ShardNumber, ShardStripeSize};
@@ -321,7 +320,7 @@ pub struct TenantConf {
     pub compaction_period: Duration,
     // Level0 delta layer threshold for compaction.
     pub compaction_threshold: usize,
-    pub compaction_algorithm: CompactionAlgorithmSettings,
+    pub compaction_algorithm: CompactionAlgorithm,
     // Determines how much history is retained, to allow
     // branching and read replicas at an older point in time.
     // The unit is #of bytes of WAL.
@@ -374,8 +373,6 @@ pub struct TenantConf {
 
     /// Switch to a new aux file policy. Switching this flag requires the user has not written any aux file into
     /// the storage before, and this flag cannot be switched back. Otherwise there will be data corruptions.
-    /// There is a `last_aux_file_policy` flag which gets persisted in `index_part.json` once the first aux
-    /// file is written.
     pub switch_aux_file_policy: AuxFilePolicy,
 }
 
@@ -407,7 +404,7 @@ pub struct TenantConfOpt {
 
     #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(default)]
-    pub compaction_algorithm: Option<CompactionAlgorithmSettings>,
+    pub compaction_algorithm: Option<CompactionAlgorithm>,
 
     #[serde(skip_serializing_if = "Option::is_none")]
     #[serde(default)]
@@ -498,9 +495,7 @@ impl TenantConfOpt {
                 .unwrap_or(global_conf.compaction_threshold),
             compaction_algorithm: self
                 .compaction_algorithm
-                .as_ref()
-                .unwrap_or(&global_conf.compaction_algorithm)
-                .clone(),
+                .unwrap_or(global_conf.compaction_algorithm),
             gc_horizon: self.gc_horizon.unwrap_or(global_conf.gc_horizon),
             gc_period: self.gc_period.unwrap_or(global_conf.gc_period),
             image_creation_threshold: self
@@ -553,9 +548,7 @@ impl Default for TenantConf {
             compaction_period: humantime::parse_duration(DEFAULT_COMPACTION_PERIOD)
                 .expect("cannot parse default compaction period"),
             compaction_threshold: DEFAULT_COMPACTION_THRESHOLD,
-            compaction_algorithm: CompactionAlgorithmSettings {
-                kind: DEFAULT_COMPACTION_ALGORITHM,
-            },
+            compaction_algorithm: DEFAULT_COMPACTION_ALGORITHM,
             gc_horizon: DEFAULT_GC_HORIZON,
             gc_period: humantime::parse_duration(DEFAULT_GC_PERIOD)
                 .expect("cannot parse default gc period"),
@@ -581,7 +574,7 @@ impl Default for TenantConf {
             lazy_slru_download: false,
             timeline_get_throttle: crate::tenant::throttle::Config::disabled(),
             image_layer_creation_check_threshold: DEFAULT_IMAGE_LAYER_CREATION_CHECK_THRESHOLD,
-            switch_aux_file_policy: AuxFilePolicy::default_tenant_config(),
+            switch_aux_file_policy: AuxFilePolicy::V1,
         }
     }
 }

pageserver/src/tenant/mgr.rs

@@ -7,7 +7,7 @@ use itertools::Itertools;
 use pageserver_api::key::Key;
 use pageserver_api::models::LocationConfigMode;
 use pageserver_api::shard::{
-    ShardCount, ShardIdentity, ShardIndex, ShardNumber, ShardStripeSize, TenantShardId,
+    ShardCount, ShardIdentity, ShardNumber, ShardStripeSize, TenantShardId,
 };
 use pageserver_api::upcall_api::ReAttachResponseTenant;
 use rand::{distributions::Alphanumeric, Rng};
@@ -127,8 +127,6 @@ pub(crate) enum ShardSelector {
     First,
     /// Pick the shard that holds this key
     Page(Key),
-    /// The shard ID is known: pick the given shard
-    Known(ShardIndex),
 }
 
 /// A convenience for use with the re_attach ControlPlaneClient function: rather
@@ -2069,11 +2067,6 @@ impl TenantManager {
                                 return ShardResolveResult::Found(tenant.clone());
                             }
                         }
-                        ShardSelector::Known(shard)
-                            if tenant.shard_identity.shard_index() == shard =>
-                        {
-                            return ShardResolveResult::Found(tenant.clone());
-                        }
                         _ => continue,
                     }
                 }

pageserver/src/tenant/remote_timeline_client.rs

@@ -189,7 +189,6 @@ use camino::Utf8Path;
 use chrono::{NaiveDateTime, Utc};
 
 pub(crate) use download::download_initdb_tar_zst;
-use pageserver_api::models::AuxFilePolicy;
 use pageserver_api::shard::{ShardIndex, TenantShardId};
 use scopeguard::ScopeGuard;
 use tokio_util::sync::CancellationToken;
@@ -519,7 +518,6 @@ impl RemoteTimelineClient {
         &self,
         layer_file_name: &LayerName,
         layer_metadata: &LayerFileMetadata,
-        local_path: &Utf8Path,
         cancel: &CancellationToken,
         ctx: &RequestContext,
     ) -> anyhow::Result<u64> {
@@ -538,7 +536,6 @@ impl RemoteTimelineClient {
                 self.timeline_id,
                 layer_file_name,
                 layer_metadata,
-                local_path,
                 cancel,
                 ctx,
             )
@@ -612,17 +609,6 @@ impl RemoteTimelineClient {
         Ok(())
     }
 
-    /// Launch an index-file upload operation in the background, with only aux_file_policy flag updated.
-    pub(crate) fn schedule_index_upload_for_aux_file_policy_update(
-        self: &Arc<Self>,
-        last_aux_file_policy: Option<AuxFilePolicy>,
-    ) -> anyhow::Result<()> {
-        let mut guard = self.upload_queue.lock().unwrap();
-        let upload_queue = guard.initialized_mut()?;
-        upload_queue.last_aux_file_policy = last_aux_file_policy;
-        self.schedule_index_upload(upload_queue);
-        Ok(())
-    }
     ///
     /// Launch an index-file upload operation in the background, if necessary.
     ///
@@ -1863,7 +1849,6 @@ impl RemoteTimelineClient {
                         dangling_files: HashMap::default(),
                         shutting_down: false,
                         shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
-                        last_aux_file_policy: initialized.last_aux_file_policy,
                     };
 
                     let upload_queue = std::mem::replace(

pageserver/src/tenant/remote_timeline_client/download.rs

@@ -21,6 +21,7 @@ use crate::config::PageServerConf;
 use crate::context::RequestContext;
 use crate::span::debug_assert_current_span_has_tenant_and_timeline_id;
 use crate::tenant::remote_timeline_client::{remote_layer_path, remote_timelines_path};
+use crate::tenant::storage_layer::layer::local_layer_path;
 use crate::tenant::storage_layer::LayerName;
 use crate::tenant::Generation;
 use crate::virtual_file::{on_fatal_io_error, MaybeFatalIo, VirtualFile};
@@ -49,13 +50,19 @@ pub async fn download_layer_file<'a>(
     timeline_id: TimelineId,
     layer_file_name: &'a LayerName,
     layer_metadata: &'a LayerFileMetadata,
-    local_path: &Utf8Path,
     cancel: &CancellationToken,
     ctx: &RequestContext,
 ) -> Result<u64, DownloadError> {
     debug_assert_current_span_has_tenant_and_timeline_id();
 
     let timeline_path = conf.timeline_path(&tenant_shard_id, &timeline_id);
+    let local_path = local_layer_path(
+        conf,
+        &tenant_shard_id,
+        &timeline_id,
+        layer_file_name,
+        &layer_metadata.generation,
+    );
 
     let remote_path = remote_layer_path(
         &tenant_shard_id.tenant_id,
@@ -75,7 +82,7 @@ pub async fn download_layer_file<'a>(
     // For more context about durable_rename check this email from postgres mailing list:
     // https://www.postgresql.org/message-id/56583BDD.9060302@2ndquadrant.com
     // If pageserver crashes the temp file will be deleted on startup and re-downloaded.
-    let temp_file_path = path_with_suffix_extension(local_path, TEMP_DOWNLOAD_EXTENSION);
+    let temp_file_path = path_with_suffix_extension(&local_path, TEMP_DOWNLOAD_EXTENSION);
 
     let bytes_amount = download_retry(
         || async { download_object(storage, &remote_path, &temp_file_path, cancel, ctx).await },

pageserver/src/tenant/remote_timeline_client/index.rs

@@ -5,7 +5,6 @@
 use std::collections::HashMap;
 
 use chrono::NaiveDateTime;
-use pageserver_api::models::AuxFilePolicy;
 use serde::{Deserialize, Serialize};
 use utils::id::TimelineId;
 
@@ -89,16 +88,6 @@ pub struct IndexPart {
 
     #[serde(default)]
     pub(crate) lineage: Lineage,
-
-    /// Describes the kind of aux files stored in the timeline.
-    ///
-    /// The value is modified during file ingestion when the latest wanted value communicated via tenant config is applied if it is acceptable.
-    /// A V1 setting after V2 files have been committed is not accepted.
-    ///
-    /// None means no aux files have been written to the storage before the point
-    /// when this flag is introduced.
-    #[serde(skip_serializing_if = "Option::is_none", default)]
-    pub(crate) last_aux_file_policy: Option<AuxFilePolicy>,
 }
 
 impl IndexPart {
@@ -112,11 +101,10 @@ impl IndexPart {
     ///      is always generated from the keys of `layer_metadata`)
     /// - 4: timeline_layers is fully removed.
     /// - 5: lineage was added
-    /// - 6: last_aux_file_policy is added.
-    const LATEST_VERSION: usize = 6;
+    const LATEST_VERSION: usize = 5;
 
     // Versions we may see when reading from a bucket.
-    pub const KNOWN_VERSIONS: &'static [usize] = &[1, 2, 3, 4, 5, 6];
+    pub const KNOWN_VERSIONS: &'static [usize] = &[1, 2, 3, 4, 5];
 
     pub const FILE_NAME: &'static str = "index_part.json";
 
@@ -125,7 +113,6 @@ impl IndexPart {
         disk_consistent_lsn: Lsn,
         metadata: TimelineMetadata,
         lineage: Lineage,
-        last_aux_file_policy: Option<AuxFilePolicy>,
     ) -> Self {
         let layer_metadata = layers_and_metadata
             .iter()
@@ -139,7 +126,6 @@ impl IndexPart {
             metadata,
             deleted_at: None,
             lineage,
-            last_aux_file_policy,
         }
     }
 
@@ -169,13 +155,8 @@ impl IndexPart {
             example_metadata.disk_consistent_lsn(),
             example_metadata,
             Default::default(),
-            Some(AuxFilePolicy::V1),
         )
     }
-
-    pub(crate) fn last_aux_file_policy(&self) -> Option<AuxFilePolicy> {
-        self.last_aux_file_policy
-    }
 }
 
 impl From<&UploadQueueInitialized> for IndexPart {
@@ -184,13 +165,7 @@ impl From<&UploadQueueInitialized> for IndexPart {
         let metadata = uq.latest_metadata.clone();
         let lineage = uq.latest_lineage.clone();
 
-        Self::new(
-            &uq.latest_files,
-            disk_consistent_lsn,
-            metadata,
-            lineage,
-            uq.last_aux_file_policy,
-        )
+        Self::new(&uq.latest_files, disk_consistent_lsn, metadata, lineage)
     }
 }
 
@@ -324,7 +299,6 @@ mod tests {
             metadata: TimelineMetadata::from_bytes(&[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
             deleted_at: None,
             lineage: Lineage::default(),
-            last_aux_file_policy: None,
         };
 
         let part = IndexPart::from_s3_bytes(example.as_bytes()).unwrap();
@@ -366,7 +340,6 @@ mod tests {
             metadata: TimelineMetadata::from_bytes(&[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
             deleted_at: None,
             lineage: Lineage::default(),
-            last_aux_file_policy: None,
         };
 
         let part = IndexPart::from_s3_bytes(example.as_bytes()).unwrap();
@@ -410,7 +383,6 @@ mod tests {
             deleted_at: Some(chrono::NaiveDateTime::parse_from_str(
                 "2023-07-31T09:00:00.123000000", "%Y-%m-%dT%H:%M:%S.%f").unwrap()),
             lineage: Lineage::default(),
-            last_aux_file_policy: None,
         };
 
         let part = IndexPart::from_s3_bytes(example.as_bytes()).unwrap();
@@ -456,7 +428,6 @@ mod tests {
             .unwrap(),
             deleted_at: None,
             lineage: Lineage::default(),
-            last_aux_file_policy: None,
         };
 
         let empty_layers_parsed = IndexPart::from_s3_bytes(empty_layers_json.as_bytes()).unwrap();
@@ -497,7 +468,6 @@ mod tests {
             metadata: TimelineMetadata::from_bytes(&[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
             deleted_at: Some(parse_naive_datetime("2023-07-31T09:00:00.123000000")),
             lineage: Lineage::default(),
-            last_aux_file_policy: None,
         };
 
         let part = IndexPart::from_s3_bytes(example.as_bytes()).unwrap();
@@ -541,57 +511,6 @@ mod tests {
                 reparenting_history: vec![TimelineId::from_str("e1bfd8c633d713d279e6fcd2bcc15b6d").unwrap()],
                 original_ancestor: Some((TimelineId::from_str("e2bfd8c633d713d279e6fcd2bcc15b6d").unwrap(), Lsn::from_str("0/15A7618").unwrap(), parse_naive_datetime("2024-05-07T18:52:36.322426563"))),
             },
-            last_aux_file_policy: None,
-        };
-
-        let part = IndexPart::from_s3_bytes(example.as_bytes()).unwrap();
-        assert_eq!(part, expected);
-    }
-
-    #[test]
-    fn v6_indexpart_is_parsed() {
-        let example = r#"{
-            "version":6,
-            "layer_metadata":{
-                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9": { "file_size": 25600000 },
-                "000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51": { "file_size": 9007199254741001 }
-            },
-            "disk_consistent_lsn":"0/16960E8",
-            "metadata_bytes":[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0],
-            "deleted_at": "2023-07-31T09:00:00.123",
-            "lineage":{
-                "original_ancestor":["e2bfd8c633d713d279e6fcd2bcc15b6d","0/15A7618","2024-05-07T18:52:36.322426563"],
-                "reparenting_history":["e1bfd8c633d713d279e6fcd2bcc15b6d"]
-            },
-            "last_aux_file_policy": "V2"
-        }"#;
-
-        let expected = IndexPart {
-            version: 6,
-            layer_metadata: HashMap::from([
-                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__0000000001696070-00000000016960E9".parse().unwrap(), IndexLayerMetadata {
-                    file_size: 25600000,
-                    generation: Generation::none(),
-                    shard: ShardIndex::unsharded()
-                }),
-                ("000000000000000000000000000000000000-FFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFFF__00000000016B59D8-00000000016B5A51".parse().unwrap(), IndexLayerMetadata {
-                    // serde_json should always parse this but this might be a double with jq for
-                    // example.
-                    file_size: 9007199254741001,
-                    generation: Generation::none(),
-                    shard: ShardIndex::unsharded()
-                })
-            ]),
-            disk_consistent_lsn: "0/16960E8".parse::<Lsn>().unwrap(),
-            metadata: TimelineMetadata::from_bytes(&[113,11,159,210,0,54,0,4,0,0,0,0,1,105,96,232,1,0,0,0,0,1,105,96,112,0,0,0,0,0,0,0,0,0,0,0,0,0,1,105,96,112,0,0,0,0,1,105,96,112,0,0,0,14,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]).unwrap(),
-            deleted_at: Some(chrono::NaiveDateTime::parse_from_str(
-                "2023-07-31T09:00:00.123000000", "%Y-%m-%dT%H:%M:%S.%f").unwrap()),
-            lineage: Lineage {
-                reparenting_history_truncated: false,
-                reparenting_history: vec![TimelineId::from_str("e1bfd8c633d713d279e6fcd2bcc15b6d").unwrap()],
-                original_ancestor: Some((TimelineId::from_str("e2bfd8c633d713d279e6fcd2bcc15b6d").unwrap(), Lsn::from_str("0/15A7618").unwrap(), parse_naive_datetime("2024-05-07T18:52:36.322426563"))),
-            },
-            last_aux_file_policy: Some(AuxFilePolicy::V2),
         };
 
         let part = IndexPart::from_s3_bytes(example.as_bytes()).unwrap();

pageserver/src/tenant/secondary.rs

@@ -6,9 +6,11 @@ mod scheduler;
 use std::{sync::Arc, time::SystemTime};
 
 use crate::{
+    config::PageServerConf,
     context::RequestContext,
     disk_usage_eviction_task::DiskUsageEvictionInfo,
     task_mgr::{self, TaskKind, BACKGROUND_RUNTIME},
+    virtual_file::MaybeFatalIo,
 };
 
 use self::{
@@ -19,8 +21,9 @@ use self::{
 use super::{
     config::{SecondaryLocationConfig, TenantConfOpt},
     mgr::TenantManager,
+    remote_timeline_client::LayerFileMetadata,
     span::debug_assert_current_span_has_tenant_id,
-    storage_layer::LayerName,
+    storage_layer::{layer::local_layer_path, LayerName},
 };
 
 use pageserver_api::{
@@ -175,7 +178,13 @@ impl SecondaryTenant {
 
     /// Cancellation safe, but on cancellation the eviction will go through
     #[instrument(skip_all, fields(tenant_id=%self.tenant_shard_id.tenant_id, shard_id=%self.tenant_shard_id.shard_slug(), timeline_id=%timeline_id, name=%name))]
-    pub(crate) async fn evict_layer(self: &Arc<Self>, timeline_id: TimelineId, name: LayerName) {
+    pub(crate) async fn evict_layer(
+        self: &Arc<Self>,
+        conf: &PageServerConf,
+        timeline_id: TimelineId,
+        name: LayerName,
+        metadata: LayerFileMetadata,
+    ) {
         debug_assert_current_span_has_tenant_id();
 
         let guard = match self.gate.enter() {
@@ -188,11 +197,41 @@ impl SecondaryTenant {
 
         let now = SystemTime::now();
 
+        let local_path = local_layer_path(
+            conf,
+            &self.tenant_shard_id,
+            &timeline_id,
+            &name,
+            &metadata.generation,
+        );
+
         let this = self.clone();
 
         // spawn it to be cancellation safe
         tokio::task::spawn_blocking(move || {
             let _guard = guard;
+            // We tolerate ENOENT, because between planning eviction and executing
+            // it, the secondary downloader could have seen an updated heatmap that
+            // resulted in a layer being deleted.
+            // Other local I/O errors are process-fatal: these should never happen.
+            let deleted = std::fs::remove_file(local_path);
+
+            let not_found = deleted
+                .as_ref()
+                .is_err_and(|x| x.kind() == std::io::ErrorKind::NotFound);
+
+            let deleted = if not_found {
+                false
+            } else {
+                deleted
+                    .map(|()| true)
+                    .fatal_err("Deleting layer during eviction")
+            };
+
+            if !deleted {
+                // skip updating accounting and putting perhaps later timestamp
+                return;
+            }
 
             // Update the timeline's state.  This does not have to be synchronized with
             // the download process, because:
@@ -211,15 +250,8 @@ impl SecondaryTenant {
             // of the cache.
             let mut detail = this.detail.lock().unwrap();
             if let Some(timeline_detail) = detail.timelines.get_mut(&timeline_id) {
-                let removed = timeline_detail.on_disk_layers.remove(&name);
-
-                // We might race with removal of the same layer during downloads, if it was removed
-                // from the heatmap.  If we see that the OnDiskState is gone, then no need to
-                // do a physical deletion or store in evicted_at.
-                if let Some(removed) = removed {
-                    removed.remove_blocking();
-                    timeline_detail.evicted_at.insert(name, now);
-                }
+                timeline_detail.on_disk_layers.remove(&name);
+                timeline_detail.evicted_at.insert(name, now);
             }
         })
         .await

pageserver/src/tenant/secondary/downloader.rs

@@ -62,10 +62,14 @@ use super::{
     CommandRequest, DownloadCommand,
 };
 
-/// For each tenant, default period for how long must have passed since the last download_tenant call before
-/// calling it again.  This default is replaced with the value of [`HeatMapTenant::upload_period_ms`] after first
-/// download, if the uploader populated it.
-const DEFAULT_DOWNLOAD_INTERVAL: Duration = Duration::from_millis(60000);
+/// For each tenant, how long must have passed since the last download_tenant call before
+/// calling it again.  This is approximately the time by which local data is allowed
+/// to fall behind remote data.
+///
+/// TODO: this should just be a default, and the actual period should be controlled
+/// via the heatmap itself
+/// `<ttps://github.com/neondatabase/neon/issues/6200>`
+const DOWNLOAD_FRESHEN_INTERVAL: Duration = Duration::from_millis(60000);
 
 /// Range of concurrency we may use when downloading layers within a timeline.  This is independent
 /// for each tenant we're downloading: the concurrency of _tenants_ is defined separately in
@@ -93,7 +97,7 @@ pub(super) async fn downloader_task(
 
     scheduler
         .run(command_queue, background_jobs_can_start, cancel)
-        .instrument(info_span!("secondary_download_scheduler"))
+        .instrument(info_span!("secondary_downloads"))
         .await
 }
 
@@ -107,7 +111,6 @@ struct SecondaryDownloader {
 pub(super) struct OnDiskState {
     metadata: LayerFileMetadata,
     access_time: SystemTime,
-    local_path: Utf8PathBuf,
 }
 
 impl OnDiskState {
@@ -118,26 +121,12 @@ impl OnDiskState {
         _ame: LayerName,
         metadata: LayerFileMetadata,
         access_time: SystemTime,
-        local_path: Utf8PathBuf,
     ) -> Self {
         Self {
             metadata,
             access_time,
-            local_path,
         }
     }
-
-    // This is infallible, because all errors are either acceptable (ENOENT), or totally
-    // unexpected (fatal).
-    pub(super) fn remove_blocking(&self) {
-        // We tolerate ENOENT, because between planning eviction and executing
-        // it, the secondary downloader could have seen an updated heatmap that
-        // resulted in a layer being deleted.
-        // Other local I/O errors are process-fatal: these should never happen.
-        std::fs::remove_file(&self.local_path)
-            .or_else(fs_ext::ignore_not_found)
-            .fatal_err("Deleting secondary layer")
-    }
 }
 
 #[derive(Debug, Clone, Default)]
@@ -148,22 +137,14 @@ pub(super) struct SecondaryDetailTimeline {
     pub(super) evicted_at: HashMap<LayerName, SystemTime>,
 }
 
-// Aspects of a heatmap that we remember after downloading it
-#[derive(Clone, Debug)]
-struct DownloadSummary {
-    etag: Etag,
-    #[allow(unused)]
-    mtime: SystemTime,
-    upload_period: Duration,
-}
-
 /// This state is written by the secondary downloader, it is opaque
 /// to TenantManager
 #[derive(Debug)]
 pub(super) struct SecondaryDetail {
     pub(super) config: SecondaryLocationConfig,
 
-    last_download: Option<DownloadSummary>,
+    last_download: Option<Instant>,
+    last_etag: Option<Etag>,
     next_download: Option<Instant>,
     pub(super) timelines: HashMap<TimelineId, SecondaryDetailTimeline>,
 }
@@ -193,6 +174,7 @@ impl SecondaryDetail {
         Self {
             config,
             last_download: None,
+            last_etag: None,
             next_download: None,
             timelines: HashMap::new(),
         }
@@ -246,8 +228,9 @@ impl SecondaryDetail {
 
 struct PendingDownload {
     secondary_state: Arc<SecondaryTenant>,
-    last_download: Option<DownloadSummary>,
+    last_download: Option<Instant>,
     target_time: Option<Instant>,
+    period: Option<Duration>,
 }
 
 impl scheduler::PendingJob for PendingDownload {
@@ -297,17 +280,10 @@ impl JobGenerator<PendingDownload, RunningDownload, CompleteDownload, DownloadCo
 
         tracing::debug!("Secondary tenant download completed");
 
+        // Update freshened_at even if there was an error: we don't want errored tenants to implicitly
+        // take priority to run again.
         let mut detail = secondary_state.detail.lock().unwrap();
-
-        let period = detail
-            .last_download
-            .as_ref()
-            .map(|d| d.upload_period)
-            .unwrap_or(DEFAULT_DOWNLOAD_INTERVAL);
-
-        // We advance next_download irrespective of errors: we don't want error cases to result in
-        // expensive busy-polling.
-        detail.next_download = Some(Instant::now() + period_jitter(period, 5));
+        detail.next_download = Some(Instant::now() + period_jitter(DOWNLOAD_FRESHEN_INTERVAL, 5));
     }
 
     async fn schedule(&mut self) -> SchedulingResult<PendingDownload> {
@@ -340,11 +316,11 @@ impl JobGenerator<PendingDownload, RunningDownload, CompleteDownload, DownloadCo
                     if detail.next_download.is_none() {
                         // Initialize randomly in the range from 0 to our interval: this uniformly spreads the start times.  Subsequent
                         // rounds will use a smaller jitter to avoid accidentally synchronizing later.
-                        detail.next_download = Some(now.checked_add(period_warmup(DEFAULT_DOWNLOAD_INTERVAL)).expect(
+                        detail.next_download = Some(now.checked_add(period_warmup(DOWNLOAD_FRESHEN_INTERVAL)).expect(
                         "Using our constant, which is known to be small compared with clock range",
                     ));
                     }
-                    (detail.last_download.clone(), detail.next_download.unwrap())
+                    (detail.last_download, detail.next_download.unwrap())
                 };
 
                 if now > next_download {
@@ -352,6 +328,7 @@ impl JobGenerator<PendingDownload, RunningDownload, CompleteDownload, DownloadCo
                         secondary_state: secondary_tenant,
                         last_download,
                         target_time: Some(next_download),
+                        period: Some(DOWNLOAD_FRESHEN_INTERVAL),
                     })
                 } else {
                     None
@@ -377,6 +354,7 @@ impl JobGenerator<PendingDownload, RunningDownload, CompleteDownload, DownloadCo
 
         Ok(PendingDownload {
             target_time: None,
+            period: None,
             last_download: None,
             secondary_state: tenant,
         })
@@ -393,6 +371,7 @@ impl JobGenerator<PendingDownload, RunningDownload, CompleteDownload, DownloadCo
             secondary_state,
             last_download,
             target_time,
+            period,
         } = job;
 
         let (completion, barrier) = utils::completion::channel();
@@ -429,15 +408,20 @@ impl JobGenerator<PendingDownload, RunningDownload, CompleteDownload, DownloadCo
 
             // If the job had a target execution time, we may check our final execution
             // time against that for observability purposes.
-            if let (Some(target_time), Some(last_download)) = (target_time, last_download) {
-                // Elapsed time includes any scheduling lag as well as the execution of the job
-                let elapsed = Instant::now().duration_since(target_time);
+            if let (Some(target_time), Some(period)) = (target_time, period) {
+                // Only track execution lag if this isn't our first download: otherwise, it is expected
+                // that execution will have taken longer than our configured interval, for example
+                // when starting up a pageserver and
+                if last_download.is_some() {
+                    // Elapsed time includes any scheduling lag as well as the execution of the job
+                    let elapsed = Instant::now().duration_since(target_time);
 
-                warn_when_period_overrun(
-                    elapsed,
-                    last_download.upload_period,
-                    BackgroundLoopKind::SecondaryDownload,
-                );
+                    warn_when_period_overrun(
+                        elapsed,
+                        period,
+                        BackgroundLoopKind::SecondaryDownload,
+                    );
+                }
             }
 
             CompleteDownload {
@@ -526,12 +510,12 @@ impl<'a> TenantDownloader<'a> {
         let tenant_shard_id = self.secondary_state.get_tenant_shard_id();
 
         // We will use the etag from last successful download to make the download conditional on changes
-        let last_download = self
+        let last_etag = self
             .secondary_state
             .detail
             .lock()
             .unwrap()
-            .last_download
+            .last_etag
             .clone();
 
         // Download the tenant's heatmap
@@ -540,7 +524,7 @@ impl<'a> TenantDownloader<'a> {
             etag: heatmap_etag,
             bytes: heatmap_bytes,
         } = match tokio::select!(
-            bytes = self.download_heatmap(last_download.as_ref().map(|d| &d.etag)) => {bytes?},
+            bytes = self.download_heatmap(last_etag.as_ref()) => {bytes?},
             _ = self.secondary_state.cancel.cancelled() => return Ok(())
         ) {
             HeatMapDownload::Unmodified => {
@@ -569,39 +553,6 @@ impl<'a> TenantDownloader<'a> {
             heatmap.timelines.len()
         );
 
-        // Get or initialize the local disk state for the timelines we will update
-        let mut timeline_states = HashMap::new();
-        for timeline in &heatmap.timelines {
-            let timeline_state = self
-                .secondary_state
-                .detail
-                .lock()
-                .unwrap()
-                .timelines
-                .get(&timeline.timeline_id)
-                .cloned();
-
-            let timeline_state = match timeline_state {
-                Some(t) => t,
-                None => {
-                    // We have no existing state: need to scan local disk for layers first.
-                    let timeline_state =
-                        init_timeline_state(self.conf, tenant_shard_id, timeline).await;
-
-                    // Re-acquire detail lock now that we're done with async load from local FS
-                    self.secondary_state
-                        .detail
-                        .lock()
-                        .unwrap()
-                        .timelines
-                        .insert(timeline.timeline_id, timeline_state.clone());
-                    timeline_state
-                }
-            };
-
-            timeline_states.insert(timeline.timeline_id, timeline_state);
-        }
-
         // Clean up any local layers that aren't in the heatmap.  We do this first for all timelines, on the general
         // principle that deletions should be done before writes wherever possible, and so that we can use this
         // phase to initialize our SecondaryProgress.
@@ -612,10 +563,6 @@ impl<'a> TenantDownloader<'a> {
 
         // Download the layers in the heatmap
         for timeline in heatmap.timelines {
-            let timeline_state = timeline_states
-                .remove(&timeline.timeline_id)
-                .expect("Just populated above");
-
             if self.secondary_state.cancel.is_cancelled() {
                 tracing::debug!(
                     "Cancelled before downloading timeline {}",
@@ -625,7 +572,7 @@ impl<'a> TenantDownloader<'a> {
             }
 
             let timeline_id = timeline.timeline_id;
-            self.download_timeline(timeline, timeline_state, ctx)
+            self.download_timeline(timeline, ctx)
                 .instrument(tracing::info_span!(
                     "secondary_download_timeline",
                     tenant_id=%tenant_shard_id.tenant_id,
@@ -637,30 +584,7 @@ impl<'a> TenantDownloader<'a> {
 
         // Only update last_etag after a full successful download: this way will not skip
         // the next download, even if the heatmap's actual etag is unchanged.
-        self.secondary_state.detail.lock().unwrap().last_download = Some(DownloadSummary {
-            etag: heatmap_etag,
-            mtime: heatmap_mtime,
-            upload_period: heatmap
-                .upload_period_ms
-                .map(|ms| Duration::from_millis(ms as u64))
-                .unwrap_or(DEFAULT_DOWNLOAD_INTERVAL),
-        });
-
-        // Robustness: we should have updated progress properly, but in case we didn't, make sure
-        // we don't leave the tenant in a state where we claim to have successfully downloaded
-        // everything, but our progress is incomplete.  The invariant here should be that if
-        // we have set `last_download` to this heatmap's etag, then the next time we see that
-        // etag we can safely do no work (i.e. we must be complete).
-        let mut progress = self.secondary_state.progress.lock().unwrap();
-        debug_assert!(progress.layers_downloaded == progress.layers_total);
-        debug_assert!(progress.bytes_downloaded == progress.bytes_total);
-        if progress.layers_downloaded != progress.layers_total
-            || progress.bytes_downloaded != progress.bytes_total
-        {
-            tracing::warn!("Correcting drift in progress stats ({progress:?})");
-            progress.layers_downloaded = progress.layers_total;
-            progress.bytes_downloaded = progress.bytes_total;
-        }
+        self.secondary_state.detail.lock().unwrap().last_etag = Some(heatmap_etag);
 
         Ok(())
     }
@@ -837,7 +761,6 @@ impl<'a> TenantDownloader<'a> {
     async fn download_timeline(
         &self,
         timeline: HeatMapTimeline,
-        timeline_state: SecondaryDetailTimeline,
         ctx: &RequestContext,
     ) -> Result<(), UpdateError> {
         debug_assert_current_span_has_tenant_and_timeline_id();
@@ -846,6 +769,34 @@ impl<'a> TenantDownloader<'a> {
         // Accumulate updates to the state
         let mut touched = Vec::new();
 
+        // Clone a view of what layers already exist on disk
+        let timeline_state = self
+            .secondary_state
+            .detail
+            .lock()
+            .unwrap()
+            .timelines
+            .get(&timeline.timeline_id)
+            .cloned();
+
+        let timeline_state = match timeline_state {
+            Some(t) => t,
+            None => {
+                // We have no existing state: need to scan local disk for layers first.
+                let timeline_state =
+                    init_timeline_state(self.conf, tenant_shard_id, &timeline).await;
+
+                // Re-acquire detail lock now that we're done with async load from local FS
+                self.secondary_state
+                    .detail
+                    .lock()
+                    .unwrap()
+                    .timelines
+                    .insert(timeline.timeline_id, timeline_state.clone());
+                timeline_state
+            }
+        };
+
         tracing::debug!(timeline_id=%timeline.timeline_id, "Downloading layers, {} in heatmap", timeline.layers.len());
 
         let mut download_futs = Vec::new();
@@ -865,12 +816,20 @@ impl<'a> TenantDownloader<'a> {
                 if cfg!(debug_assertions) {
                     // Debug for https://github.com/neondatabase/neon/issues/6966: check that the files we think
                     // are already present on disk are really there.
-                    match tokio::fs::metadata(&on_disk.local_path).await {
+                    let local_path = local_layer_path(
+                        self.conf,
+                        tenant_shard_id,
+                        &timeline.timeline_id,
+                        &layer.name,
+                        &layer.metadata.generation,
+                    );
+
+                    match tokio::fs::metadata(&local_path).await {
                         Ok(meta) => {
                             tracing::debug!(
                                 "Layer {} present at {}, size {}",
                                 layer.name,
-                                on_disk.local_path,
+                                local_path,
                                 meta.len(),
                             );
                         }
@@ -878,7 +837,7 @@ impl<'a> TenantDownloader<'a> {
                             tracing::warn!(
                                 "Layer {} not found at {} ({})",
                                 layer.name,
-                                on_disk.local_path,
+                                local_path,
                                 e
                             );
                             debug_assert!(false);
@@ -967,13 +926,6 @@ impl<'a> TenantDownloader<'a> {
                         v.get_mut().access_time = t.access_time;
                     }
                     Entry::Vacant(e) => {
-                        let local_path = local_layer_path(
-                            self.conf,
-                            tenant_shard_id,
-                            &timeline.timeline_id,
-                            &t.name,
-                            &t.metadata.generation,
-                        );
                         e.insert(OnDiskState::new(
                             self.conf,
                             tenant_shard_id,
@@ -981,7 +933,6 @@ impl<'a> TenantDownloader<'a> {
                             t.name,
                             LayerFileMetadata::from(&t.metadata),
                             t.access_time,
-                            local_path,
                         ));
                     }
                 }
@@ -1004,20 +955,7 @@ impl<'a> TenantDownloader<'a> {
             &self.secondary_state.cancel
         );
 
-        let local_path = local_layer_path(
-            self.conf,
-            tenant_shard_id,
-            timeline_id,
-            &layer.name,
-            &layer.metadata.generation,
-        );
-
         // Note: no backoff::retry wrapper here because download_layer_file does its own retries internally
-        tracing::info!(
-            "Starting download of layer {}, size {}",
-            layer.name,
-            layer.metadata.file_size
-        );
         let downloaded_bytes = match download_layer_file(
             self.conf,
             self.remote_storage,
@@ -1025,7 +963,6 @@ impl<'a> TenantDownloader<'a> {
             *timeline_id,
             &layer.name,
             &LayerFileMetadata::from(&layer.metadata),
-            &local_path,
             &self.secondary_state.cancel,
             ctx,
         )
@@ -1040,14 +977,6 @@ impl<'a> TenantDownloader<'a> {
                     "Skipped downloading missing layer {}, raced with compaction/gc?",
                     layer.name
                 );
-
-                // If the layer is 404, adjust the progress statistics to reflect that we will not download it.
-                let mut progress = self.secondary_state.progress.lock().unwrap();
-                progress.layers_total = progress.layers_total.saturating_sub(1);
-                progress.bytes_total = progress
-                    .bytes_total
-                    .saturating_sub(layer.metadata.file_size);
-
                 return Ok(None);
             }
             Err(e) => return Err(e.into()),
@@ -1187,7 +1116,6 @@ async fn init_timeline_state(
                                     name,
                                     LayerFileMetadata::from(&remote_meta.metadata),
                                     remote_meta.access_time,
-                                    file_path,
                                 ),
                             );
                         }

pageserver/src/tenant/secondary/heatmap_uploader.rs

@@ -53,7 +53,7 @@ pub(super) async fn heatmap_uploader_task(
 
     scheduler
         .run(command_queue, background_jobs_can_start, cancel)
-        .instrument(info_span!("heatmap_upload_scheduler"))
+        .instrument(info_span!("heatmap_uploader"))
         .await
 }
 

pageserver/src/tenant/secondary/scheduler.rs

@@ -179,13 +179,6 @@ where
             // Schedule some work, if concurrency limit permits it
             self.spawn_pending();
 
-            // This message is printed every scheduling iteration as proof of liveness when looking at logs
-            tracing::info!(
-                "Status: {} tasks running, {} pending",
-                self.running.len(),
-                self.pending.len()
-            );
-
             // Between scheduling iterations, we will:
             //  - Drain any complete tasks and spawn pending tasks
             //  - Handle incoming administrative commands
@@ -265,11 +258,7 @@ where
 
         self.tasks.spawn(fut);
 
-        let replaced = self.running.insert(tenant_shard_id, in_progress);
-        debug_assert!(replaced.is_none());
-        if replaced.is_some() {
-            tracing::warn!(%tenant_shard_id, "Unexpectedly spawned a task when one was already running")
-        }
+        self.running.insert(tenant_shard_id, in_progress);
     }
 
     /// For all pending tenants that are elegible for execution, spawn their task.
@@ -279,9 +268,7 @@ where
         while !self.pending.is_empty() && self.running.len() < self.concurrency {
             // unwrap: loop condition includes !is_empty()
             let pending = self.pending.pop_front().unwrap();
-            if !self.running.contains_key(pending.get_tenant_shard_id()) {
-                self.do_spawn(pending);
-            }
+            self.do_spawn(pending);
         }
     }
 
@@ -334,8 +321,7 @@ where
 
         let tenant_shard_id = job.get_tenant_shard_id();
         let barrier = if let Some(barrier) = self.get_running(tenant_shard_id) {
-            tracing::info!(tenant_id=%tenant_shard_id.tenant_id, shard_id=%tenant_shard_id.shard_slug(),
-                           "Command already running, waiting for it");
+            tracing::info!("Command already running, waiting for it");
             barrier
         } else {
             let running = self.spawn_now(job);

pageserver/src/tenant/storage_layer.rs

@@ -113,20 +113,12 @@ impl From<VectoredValueReconstructState> for ValueReconstructState {
     }
 }
 
-/// Bag of data accumulated during a vectored get..
+/// Bag of data accumulated during a vectored get
 pub(crate) struct ValuesReconstructState {
-    /// The keys will be removed after `get_vectored` completes. The caller outside `Timeline`
-    /// should not expect to get anything from this hashmap.
     pub(crate) keys: HashMap<Key, Result<VectoredValueReconstructState, PageReconstructError>>,
-    /// The keys which are already retrieved
+
     keys_done: KeySpaceRandomAccum,
-
-    /// The keys covered by the image layers
-    keys_with_image_coverage: Option<Range<Key>>,
-
-    // Statistics that are still accessible as a caller of `get_vectored_impl`.
     layers_visited: u32,
-    delta_layers_visited: u32,
 }
 
 impl ValuesReconstructState {
@@ -134,9 +126,7 @@ impl ValuesReconstructState {
         Self {
             keys: HashMap::new(),
             keys_done: KeySpaceRandomAccum::new(),
-            keys_with_image_coverage: None,
             layers_visited: 0,
-            delta_layers_visited: 0,
         }
     }
 
@@ -150,17 +140,8 @@ impl ValuesReconstructState {
         }
     }
 
-    pub(crate) fn on_layer_visited(&mut self, layer: &ReadableLayer) {
+    pub(crate) fn on_layer_visited(&mut self) {
         self.layers_visited += 1;
-        if let ReadableLayer::PersistentLayer(layer) = layer {
-            if layer.layer_desc().is_delta() {
-                self.delta_layers_visited += 1;
-            }
-        }
-    }
-
-    pub(crate) fn get_delta_layers_visited(&self) -> u32 {
-        self.delta_layers_visited
     }
 
     pub(crate) fn get_layers_visited(&self) -> u32 {
@@ -190,16 +171,6 @@ impl ValuesReconstructState {
         }
     }
 
-    /// On hitting image layer, we can mark all keys in this range as done, because
-    /// if the image layer does not contain a key, it is deleted/never added.
-    pub(crate) fn on_image_layer_visited(&mut self, key_range: &Range<Key>) {
-        let prev_val = self.keys_with_image_coverage.replace(key_range.clone());
-        assert_eq!(
-            prev_val, None,
-            "should consume the keyspace before the next iteration"
-        );
-    }
-
     /// Update the state collected for a given key.
     /// Returns true if this was the last value needed for the key and false otherwise.
     ///
@@ -262,12 +233,8 @@ impl ValuesReconstructState {
 
     /// Returns the key space describing the keys that have
     /// been marked as completed since the last call to this function.
-    /// Returns individual keys done, and the image layer coverage.
-    pub(crate) fn consume_done_keys(&mut self) -> (KeySpace, Option<Range<Key>>) {
-        (
-            self.keys_done.consume_keyspace(),
-            self.keys_with_image_coverage.take(),
-        )
+    pub(crate) fn consume_done_keys(&mut self) -> KeySpace {
+        self.keys_done.consume_keyspace()
     }
 }
 

pageserver/src/tenant/storage_layer/image_layer.rs

@@ -158,7 +158,6 @@ pub struct ImageLayerInner {
     index_start_blk: u32,
     index_root_blk: u32,
 
-    key_range: Range<Key>,
     lsn: Lsn,
 
     file: VirtualFile,
@@ -420,7 +419,6 @@ impl ImageLayerInner {
             file,
             file_id,
             max_vectored_read_bytes,
-            key_range: actual_summary.key_range,
         }))
     }
 
@@ -480,8 +478,6 @@ impl ImageLayerInner {
         self.do_reads_and_update_state(reads, reconstruct_state, ctx)
             .await;
 
-        reconstruct_state.on_image_layer_visited(&self.key_range);
-
         Ok(())
     }
 
@@ -650,7 +646,7 @@ impl ImageLayerWriterInner {
                 lsn,
             },
         );
-        trace!("creating image layer {}", path);
+        info!("new image layer {path}");
         let mut file = {
             VirtualFile::open_with_options(
                 &path,
@@ -770,7 +766,7 @@ impl ImageLayerWriterInner {
         // FIXME: why not carry the virtualfile here, it supports renaming?
         let layer = Layer::finish_creating(self.conf, timeline, desc, &self.path)?;
 
-        info!("created image layer {}", layer.local_path());
+        trace!("created image layer {}", layer.local_path());
 
         Ok(layer)
     }

pageserver/src/tenant/storage_layer/layer.rs

@@ -12,7 +12,7 @@ use std::time::{Duration, SystemTime};
 use tracing::Instrument;
 use utils::id::TimelineId;
 use utils::lsn::Lsn;
-use utils::sync::{gate, heavier_once_cell};
+use utils::sync::heavier_once_cell;
 
 use crate::config::PageServerConf;
 use crate::context::{DownloadBehavior, RequestContext};
@@ -1108,7 +1108,6 @@ impl LayerInner {
             .download_layer_file(
                 &self.desc.layer_name(),
                 &self.metadata(),
-                &self.path,
                 &timeline.cancel,
                 ctx,
             )
@@ -1264,7 +1263,6 @@ impl LayerInner {
                 lsn_end: lsn_range.end,
                 remote: !resident,
                 access_stats,
-                l0: crate::tenant::layer_map::LayerMap::is_l0(self.layer_desc()),
             }
         } else {
             let lsn = self.desc.image_layer_lsn();
@@ -1333,7 +1331,7 @@ impl LayerInner {
 
         is_good_to_continue(&rx.borrow_and_update())?;
 
-        let Ok(gate) = timeline.gate.enter() else {
+        let Ok(_gate) = timeline.gate.enter() else {
             return Err(EvictionCancelled::TimelineGone);
         };
 
@@ -1421,7 +1419,7 @@ impl LayerInner {
         Self::spawn_blocking(move || {
             let _span = span.entered();
 
-            let res = self.evict_blocking(&timeline, &gate, &permit);
+            let res = self.evict_blocking(&timeline, &permit);
 
             let waiters = self.inner.initializer_count();
 
@@ -1447,7 +1445,6 @@ impl LayerInner {
     fn evict_blocking(
         &self,
         timeline: &Timeline,
-        _gate: &gate::GateGuard,
         _permit: &heavier_once_cell::InitPermit,
     ) -> Result<(), EvictionCancelled> {
         // now accesses to `self.inner.get_or_init*` wait on the semaphore or the `_permit`

pageserver/src/tenant/storage_layer/layer_name.rs

@@ -347,33 +347,37 @@ impl<'de> serde::de::Visitor<'de> for LayerNameVisitor {
 mod test {
     use super::*;
     #[test]
-    fn image_layer_parse() {
+    fn image_layer_parse() -> anyhow::Result<()> {
         let expected = LayerName::Image(ImageLayerName {
             key_range: Key::from_i128(0)
                 ..Key::from_hex("000000067F00000001000004DF0000000006").unwrap(),
             lsn: Lsn::from_hex("00000000014FED58").unwrap(),
         });
-        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-v1-00000001").unwrap();
+        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-v1-00000001").map_err(|s| anyhow::anyhow!(s))?;
         assert_eq!(parsed, expected,);
 
         // Omitting generation suffix is valid
-        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58").unwrap();
+        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58").map_err(|s| anyhow::anyhow!(s))?;
         assert_eq!(parsed, expected,);
+
+        Ok(())
     }
 
     #[test]
-    fn delta_layer_parse() {
+    fn delta_layer_parse() -> anyhow::Result<()> {
         let expected = LayerName::Delta(DeltaLayerName {
             key_range: Key::from_i128(0)
                 ..Key::from_hex("000000067F00000001000004DF0000000006").unwrap(),
             lsn_range: Lsn::from_hex("00000000014FED58").unwrap()
                 ..Lsn::from_hex("000000000154C481").unwrap(),
         });
-        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-000000000154C481-v1-00000001").unwrap();
+        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-000000000154C481-v1-00000001").map_err(|s| anyhow::anyhow!(s))?;
         assert_eq!(parsed, expected);
 
         // Omitting generation suffix is valid
-        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-000000000154C481").unwrap();
+        let parsed = LayerName::from_str("000000000000000000000000000000000000-000000067F00000001000004DF0000000006__00000000014FED58-000000000154C481").map_err(|s| anyhow::anyhow!(s))?;
         assert_eq!(parsed, expected);
+
+        Ok(())
     }
 }

pageserver/src/tenant/timeline.rs

@@ -18,14 +18,14 @@ use fail::fail_point;
 use once_cell::sync::Lazy;
 use pageserver_api::{
     key::{
-        AUX_FILES_KEY, KEY_SIZE, METADATA_KEY_BEGIN_PREFIX, METADATA_KEY_END_PREFIX,
-        NON_INHERITED_RANGE, NON_INHERITED_SPARSE_RANGE,
+        AUX_FILES_KEY, METADATA_KEY_BEGIN_PREFIX, METADATA_KEY_END_PREFIX, NON_INHERITED_RANGE,
+        NON_INHERITED_SPARSE_RANGE,
     },
-    keyspace::{KeySpaceAccum, KeySpaceRandomAccum, SparseKeyPartitioning},
+    keyspace::{KeySpaceAccum, SparseKeyPartitioning},
     models::{
-        AtomicAuxFilePolicy, AuxFilePolicy, CompactionAlgorithm, CompactionAlgorithmSettings,
-        DownloadRemoteLayersTaskInfo, DownloadRemoteLayersTaskSpawnRequest, EvictionPolicy,
-        InMemoryLayerInfo, LayerMapInfo, LsnLease, TimelineState,
+        AuxFilePolicy, CompactionAlgorithm, DownloadRemoteLayersTaskInfo,
+        DownloadRemoteLayersTaskSpawnRequest, EvictionPolicy, InMemoryLayerInfo, LayerMapInfo,
+        TimelineState,
     },
     reltag::BlockNumber,
     shard::{ShardIdentity, ShardNumber, TenantShardId},
@@ -60,6 +60,7 @@ use std::{
     ops::ControlFlow,
 };
 
+use crate::tenant::timeline::init::LocalLayerFileMetadata;
 use crate::{
     aux_file::AuxFileSizeEstimator,
     tenant::{
@@ -88,9 +89,6 @@ use crate::{
     metrics::ScanLatencyOngoingRecording, tenant::timeline::logical_size::CurrentLogicalSize,
 };
 use crate::{pgdatadir_mapping::LsnForTimestamp, tenant::tasks::BackgroundLoopKind};
-use crate::{
-    pgdatadir_mapping::MAX_AUX_FILE_V2_DELTAS, tenant::timeline::init::LocalLayerFileMetadata,
-};
 use crate::{
     pgdatadir_mapping::{AuxFilesDirectory, DirectoryKind},
     virtual_file::{MaybeFatalIo, VirtualFile},
@@ -269,8 +267,6 @@ pub struct Timeline {
     // Atomic would be more appropriate here.
     last_freeze_ts: RwLock<Instant>,
 
-    pub(crate) standby_horizon: AtomicLsn,
-
     // WAL redo manager. `None` only for broken tenants.
     walredo_mgr: Option<Arc<super::WalRedoManager>>,
 
@@ -350,8 +346,8 @@ pub struct Timeline {
     // though let's keep them both for better error visibility.
     pub initdb_lsn: Lsn,
 
-    /// When did we last calculate the partitioning? Make it pub to test cases.
-    pub(super) partitioning: tokio::sync::Mutex<((KeyPartitioning, SparseKeyPartitioning), Lsn)>,
+    /// When did we last calculate the partitioning?
+    partitioning: tokio::sync::Mutex<((KeyPartitioning, SparseKeyPartitioning), Lsn)>,
 
     /// Configuration: how often should the partitioning be recalculated.
     repartition_threshold: u64,
@@ -417,11 +413,7 @@ pub struct Timeline {
     /// Keep aux directory cache to avoid it's reconstruction on each update
     pub(crate) aux_files: tokio::sync::Mutex<AuxFilesState>,
 
-    /// Size estimator for aux file v2
     pub(crate) aux_file_size_estimator: AuxFileSizeEstimator,
-
-    /// Indicate whether aux file v2 storage is enabled.
-    pub(crate) last_aux_file_policy: AtomicAuxFilePolicy,
 }
 
 pub struct WalReceiverInfo {
@@ -485,11 +477,6 @@ impl GcCutoffs {
     }
 }
 
-pub(crate) struct TimelineVisitOutcome {
-    completed_keyspace: KeySpace,
-    image_covered_keyspace: KeySpace,
-}
-
 /// An error happened in a get() operation.
 #[derive(thiserror::Error, Debug)]
 pub(crate) enum PageReconstructError {
@@ -514,13 +501,6 @@ pub(crate) enum PageReconstructError {
     MissingKey(MissingKeyError),
 }
 
-impl GetVectoredError {
-    #[cfg(test)]
-    pub(crate) fn is_missing_key_error(&self) -> bool {
-        matches!(self, Self::MissingKey(_))
-    }
-}
-
 #[derive(Debug)]
 pub struct MissingKeyError {
     key: Key,
@@ -798,11 +778,6 @@ pub(crate) enum ShutdownMode {
     Hard,
 }
 
-struct ImageLayerCreationOutcome {
-    image: Option<ResidentLayer>,
-    next_start_key: Key,
-}
-
 /// Public interface functions
 impl Timeline {
     /// Get the LSN where this branch was created
@@ -904,7 +879,7 @@ impl Timeline {
                 }
 
                 let vectored_res = self
-                    .get_vectored_impl(keyspace.clone(), lsn, &mut reconstruct_state, ctx)
+                    .get_vectored_impl(keyspace.clone(), lsn, reconstruct_state, ctx)
                     .await;
 
                 if self.conf.validate_vectored_get {
@@ -1049,12 +1024,7 @@ impl Timeline {
             }
             GetVectoredImpl::Vectored => {
                 let vectored_res = self
-                    .get_vectored_impl(
-                        keyspace.clone(),
-                        lsn,
-                        &mut ValuesReconstructState::new(),
-                        ctx,
-                    )
+                    .get_vectored_impl(keyspace.clone(), lsn, ValuesReconstructState::new(), ctx)
                     .await;
 
                 if self.conf.validate_vectored_get {
@@ -1142,7 +1112,7 @@ impl Timeline {
             .get_vectored_impl(
                 keyspace.clone(),
                 lsn,
-                &mut ValuesReconstructState::default(),
+                ValuesReconstructState::default(),
                 ctx,
             )
             .await;
@@ -1219,7 +1189,7 @@ impl Timeline {
         &self,
         keyspace: KeySpace,
         lsn: Lsn,
-        reconstruct_state: &mut ValuesReconstructState,
+        mut reconstruct_state: ValuesReconstructState,
         ctx: &RequestContext,
     ) -> Result<BTreeMap<Key, Result<Bytes, PageReconstructError>>, GetVectoredError> {
         let get_kind = if keyspace.total_raw_size() == 1 {
@@ -1231,7 +1201,7 @@ impl Timeline {
         let get_data_timer = crate::metrics::GET_RECONSTRUCT_DATA_TIME
             .for_get_kind(get_kind)
             .start_timer();
-        self.get_vectored_reconstruct_data(keyspace, lsn, reconstruct_state, ctx)
+        self.get_vectored_reconstruct_data(keyspace, lsn, &mut reconstruct_state, ctx)
             .await?;
         get_data_timer.stop_and_record();
 
@@ -1240,8 +1210,7 @@ impl Timeline {
             .start_timer();
         let mut results: BTreeMap<Key, Result<Bytes, PageReconstructError>> = BTreeMap::new();
         let layers_visited = reconstruct_state.get_layers_visited();
-
-        for (key, res) in std::mem::take(&mut reconstruct_state.keys) {
+        for (key, res) in reconstruct_state.keys {
             match res {
                 Err(err) => {
                     results.insert(key, Err(err));
@@ -1532,20 +1501,6 @@ impl Timeline {
         Ok(())
     }
 
-    /// Obtains a temporary lease blocking garbage collection for the given LSN
-    pub(crate) fn make_lsn_lease(
-        &self,
-        _lsn: Lsn,
-        _ctx: &RequestContext,
-    ) -> anyhow::Result<LsnLease> {
-        const LEASE_LENGTH: Duration = Duration::from_secs(5 * 60);
-        let lease = LsnLease {
-            valid_until: SystemTime::now() + LEASE_LENGTH,
-        };
-        // TODO: dummy implementation
-        Ok(lease)
-    }
-
     /// Flush to disk all data that was written with the put_* functions
     #[instrument(skip(self), fields(tenant_id=%self.tenant_shard_id.tenant_id, shard_id=%self.tenant_shard_id.shard_slug(), timeline_id=%self.timeline_id))]
     pub(crate) async fn freeze_and_flush(&self) -> anyhow::Result<()> {
@@ -1700,7 +1655,7 @@ impl Timeline {
             return Ok(());
         }
 
-        match self.get_compaction_algorithm_settings().kind {
+        match self.get_compaction_algorithm() {
             CompactionAlgorithm::Tiered => self.compact_tiered(cancel, ctx).await,
             CompactionAlgorithm::Legacy => self.compact_legacy(cancel, flags, ctx).await,
         }
@@ -2096,14 +2051,12 @@ impl Timeline {
             .unwrap_or(self.conf.default_tenant_conf.image_creation_threshold)
     }
 
-    fn get_compaction_algorithm_settings(&self) -> CompactionAlgorithmSettings {
+    fn get_compaction_algorithm(&self) -> CompactionAlgorithm {
         let tenant_conf = &self.tenant_conf.load();
         tenant_conf
             .tenant_conf
             .compaction_algorithm
-            .as_ref()
-            .unwrap_or(&self.conf.default_tenant_conf.compaction_algorithm)
-            .clone()
+            .unwrap_or(self.conf.default_tenant_conf.compaction_algorithm)
     }
 
     fn get_eviction_policy(&self) -> EvictionPolicy {
@@ -2180,7 +2133,6 @@ impl Timeline {
         resources: TimelineResources,
         pg_version: u32,
         state: TimelineState,
-        aux_file_policy: Option<AuxFilePolicy>,
         cancel: CancellationToken,
     ) -> Arc<Self> {
         let disk_consistent_lsn = metadata.disk_consistent_lsn();
@@ -2297,8 +2249,6 @@ impl Timeline {
                 compaction_lock: tokio::sync::Mutex::default(),
                 gc_lock: tokio::sync::Mutex::default(),
 
-                standby_horizon: AtomicLsn::new(0),
-
                 timeline_get_throttle: resources.timeline_get_throttle,
 
                 aux_files: tokio::sync::Mutex::new(AuxFilesState {
@@ -2307,8 +2257,6 @@ impl Timeline {
                 }),
 
                 aux_file_size_estimator: AuxFileSizeEstimator::new(aux_file_metrics),
-
-                last_aux_file_policy: AtomicAuxFilePolicy::new(aux_file_policy),
             };
             result.repartition_threshold =
                 result.get_checkpoint_distance() / REPARTITION_FREQ_IN_CHECKPOINT_DISTANCE;
@@ -3332,15 +3280,12 @@ impl Timeline {
 
         let mut cont_lsn = Lsn(request_lsn.0 + 1);
 
-        let missing_keyspace = loop {
+        loop {
             if self.cancel.is_cancelled() {
                 return Err(GetVectoredError::Cancelled);
             }
 
-            let TimelineVisitOutcome {
-                completed_keyspace: completed,
-                image_covered_keyspace,
-            } = Self::get_vectored_reconstruct_data_timeline(
+            let completed = Self::get_vectored_reconstruct_data_timeline(
                 timeline,
                 keyspace.clone(),
                 cont_lsn,
@@ -3359,31 +3304,12 @@ impl Timeline {
                 ranges: vec![NON_INHERITED_RANGE, NON_INHERITED_SPARSE_RANGE],
             });
 
-            // Keyspace is fully retrieved
-            if keyspace.is_empty() {
-                break None;
+            // Keyspace is fully retrieved, no ancestor timeline, or metadata scan (where we do not look
+            // into ancestor timelines). TODO: is there any other metadata which we want to inherit?
+            if keyspace.total_raw_size() == 0 || timeline.ancestor_timeline.is_none() {
+                break;
             }
 
-            // Not fully retrieved but no ancestor timeline.
-            if timeline.ancestor_timeline.is_none() {
-                break Some(keyspace);
-            }
-
-            // Now we see if there are keys covered by the image layer but does not exist in the
-            // image layer, which means that the key does not exist.
-
-            // The block below will stop the vectored search if any of the keys encountered an image layer
-            // which did not contain a snapshot for said key. Since we have already removed all completed
-            // keys from `keyspace`, we expect there to be no overlap between it and the image covered key
-            // space. If that's not the case, we had at least one key encounter a gap in the image layer
-            // and stop the search as a result of that.
-            let removed = keyspace.remove_overlapping_with(&image_covered_keyspace);
-            if !removed.is_empty() {
-                break Some(removed);
-            }
-            // If we reached this point, `remove_overlapping_with` should not have made any change to the
-            // keyspace.
-
             // Take the min to avoid reconstructing a page with data newer than request Lsn.
             cont_lsn = std::cmp::min(Lsn(request_lsn.0 + 1), Lsn(timeline.ancestor_lsn.0 + 1));
             timeline_owned = timeline
@@ -3391,14 +3317,14 @@ impl Timeline {
                 .await
                 .map_err(GetVectoredError::GetReadyAncestorError)?;
             timeline = &*timeline_owned;
-        };
+        }
 
-        if let Some(missing_keyspace) = missing_keyspace {
+        if keyspace.total_raw_size() != 0 {
             return Err(GetVectoredError::MissingKey(MissingKeyError {
-                key: missing_keyspace.start().unwrap(), /* better if we can store the full keyspace */
+                key: keyspace.start().unwrap(), /* better if we can store the full keyspace */
                 shard: self
                     .shard_identity
-                    .get_shard_number(&missing_keyspace.start().unwrap()),
+                    .get_shard_number(&keyspace.start().unwrap()),
                 cont_lsn,
                 request_lsn,
                 ancestor_lsn: Some(timeline.ancestor_lsn),
@@ -3423,9 +3349,6 @@ impl Timeline {
     ///
     /// At each iteration pop the top of the fringe (the layer with the highest Lsn)
     /// and get all the required reconstruct data from the layer in one go.
-    ///
-    /// Returns the completed keyspace and the keyspaces with image coverage. The caller
-    /// decides how to deal with these two keyspaces.
     async fn get_vectored_reconstruct_data_timeline(
         timeline: &Timeline,
         keyspace: KeySpace,
@@ -3433,27 +3356,20 @@ impl Timeline {
         reconstruct_state: &mut ValuesReconstructState,
         cancel: &CancellationToken,
         ctx: &RequestContext,
-    ) -> Result<TimelineVisitOutcome, GetVectoredError> {
+    ) -> Result<KeySpace, GetVectoredError> {
         let mut unmapped_keyspace = keyspace.clone();
         let mut fringe = LayerFringe::new();
 
         let mut completed_keyspace = KeySpace::default();
-        let mut image_covered_keyspace = KeySpaceRandomAccum::new();
 
         loop {
             if cancel.is_cancelled() {
                 return Err(GetVectoredError::Cancelled);
             }
 
-            let (keys_done_last_step, keys_with_image_coverage) =
-                reconstruct_state.consume_done_keys();
+            let keys_done_last_step = reconstruct_state.consume_done_keys();
             unmapped_keyspace.remove_overlapping_with(&keys_done_last_step);
             completed_keyspace.merge(&keys_done_last_step);
-            if let Some(keys_with_image_coverage) = keys_with_image_coverage {
-                unmapped_keyspace
-                    .remove_overlapping_with(&KeySpace::single(keys_with_image_coverage.clone()));
-                image_covered_keyspace.add_range(keys_with_image_coverage);
-            }
 
             // Do not descent any further if the last layer we visited
             // completed all keys in the keyspace it inspected. This is not
@@ -3525,16 +3441,13 @@ impl Timeline {
                 unmapped_keyspace = keyspace_to_read;
                 cont_lsn = next_cont_lsn;
 
-                reconstruct_state.on_layer_visited(&layer_to_read);
+                reconstruct_state.on_layer_visited();
             } else {
                 break;
             }
         }
 
-        Ok(TimelineVisitOutcome {
-            completed_keyspace,
-            image_covered_keyspace: image_covered_keyspace.consume_keyspace(),
-        })
+        Ok(completed_keyspace)
     }
 
     /// # Cancel-safety
@@ -3882,7 +3795,7 @@ impl Timeline {
             }
         }
 
-        let (layers_to_upload, delta_layers_to_add) = if create_image_layer {
+        let (layers_to_upload, delta_layer_to_add) = if create_image_layer {
             // Note: The 'ctx' in use here has DownloadBehavior::Error. We should not
             // require downloading anything during initial import.
             let ((rel_partition, metadata_partition), _lsn) = self
@@ -3899,23 +3812,26 @@ impl Timeline {
             }
 
             // For metadata, always create delta layers.
-            let delta_layers = if !metadata_partition.parts.is_empty() {
-                // In the current implementation, the metadata partition will only have one part, and the part will only have
-                // one single key range. This might change in the future.
-                let mut delta_layers_created = Vec::new();
-                for ks in &metadata_partition.parts {
-                    for range in &ks.0.ranges {
-                        let layer = self
-                            .create_delta_layer(&frozen_layer, Some(range.clone()), ctx)
-                            .await?;
-                        if let Some(layer) = layer {
-                            delta_layers_created.push(layer);
-                        }
-                    }
-                }
-                delta_layers_created
+            let delta_layer = if !metadata_partition.parts.is_empty() {
+                assert_eq!(
+                    metadata_partition.parts.len(),
+                    1,
+                    "currently sparse keyspace should only contain a single aux file keyspace"
+                );
+                let metadata_keyspace = &metadata_partition.parts[0];
+                assert_eq!(
+                    metadata_keyspace.0.ranges.len(),
+                    1,
+                    "aux file keyspace should be a single range"
+                );
+                self.create_delta_layer(
+                    &frozen_layer,
+                    Some(metadata_keyspace.0.ranges[0].clone()),
+                    ctx,
+                )
+                .await?
             } else {
-                Vec::new()
+                None
             };
 
             // For image layers, we add them immediately into the layer map.
@@ -3930,8 +3846,12 @@ impl Timeline {
                 .await?,
             );
 
-            layers_to_upload.extend(delta_layers.iter().cloned());
-            (layers_to_upload, delta_layers)
+            if let Some(delta_layer) = delta_layer {
+                layers_to_upload.push(delta_layer.clone());
+                (layers_to_upload, Some(delta_layer))
+            } else {
+                (layers_to_upload, None)
+            }
         } else {
             // Normal case, write out a L0 delta layer file.
             // `create_delta_layer` will not modify the layer map.
@@ -3939,7 +3859,12 @@ impl Timeline {
             let Some(layer) = self.create_delta_layer(&frozen_layer, None, ctx).await? else {
                 panic!("delta layer cannot be empty if no filter is applied");
             };
-            (vec![layer.clone()], vec![layer])
+            (
+                // FIXME: even though we have a single image and single delta layer assumption
+                // we push them to vec
+                vec![layer.clone()],
+                Some(layer),
+            )
         };
 
         pausable_failpoint!("flush-layer-cancel-after-writing-layer-out-pausable");
@@ -3960,7 +3885,7 @@ impl Timeline {
                 return Err(FlushLayerError::Cancelled);
             }
 
-            guard.finish_flush_l0_layer(&delta_layers_to_add, &frozen_layer, &self.metrics);
+            guard.finish_flush_l0_layer(delta_layer_to_add.as_ref(), &frozen_layer, &self.metrics);
 
             if self.set_disk_consistent_lsn(disk_consistent_lsn) {
                 // Schedule remote uploads that will reflect our new disk_consistent_lsn
@@ -4202,176 +4127,6 @@ impl Timeline {
         false
     }
 
-    /// Create image layers for Postgres data. Assumes the caller passes a partition that is not too large,
-    /// so that at most one image layer will be produced from this function.
-    async fn create_image_layer_for_rel_blocks(
-        self: &Arc<Self>,
-        partition: &KeySpace,
-        mut image_layer_writer: ImageLayerWriter,
-        lsn: Lsn,
-        ctx: &RequestContext,
-        img_range: Range<Key>,
-        start: Key,
-    ) -> Result<ImageLayerCreationOutcome, CreateImageLayersError> {
-        let mut wrote_keys = false;
-
-        let mut key_request_accum = KeySpaceAccum::new();
-        for range in &partition.ranges {
-            let mut key = range.start;
-            while key < range.end {
-                // Decide whether to retain this key: usually we do, but sharded tenants may
-                // need to drop keys that don't belong to them.  If we retain the key, add it
-                // to `key_request_accum` for later issuing a vectored get
-                if self.shard_identity.is_key_disposable(&key) {
-                    debug!(
-                        "Dropping key {} during compaction (it belongs on shard {:?})",
-                        key,
-                        self.shard_identity.get_shard_number(&key)
-                    );
-                } else {
-                    key_request_accum.add_key(key);
-                }
-
-                let last_key_in_range = key.next() == range.end;
-                key = key.next();
-
-                // Maybe flush `key_rest_accum`
-                if key_request_accum.raw_size() >= Timeline::MAX_GET_VECTORED_KEYS
-                    || (last_key_in_range && key_request_accum.raw_size() > 0)
-                {
-                    let results = self
-                        .get_vectored(key_request_accum.consume_keyspace(), lsn, ctx)
-                        .await?;
-
-                    for (img_key, img) in results {
-                        let img = match img {
-                            Ok(img) => img,
-                            Err(err) => {
-                                // If we fail to reconstruct a VM or FSM page, we can zero the
-                                // page without losing any actual user data. That seems better
-                                // than failing repeatedly and getting stuck.
-                                //
-                                // We had a bug at one point, where we truncated the FSM and VM
-                                // in the pageserver, but the Postgres didn't know about that
-                                // and continued to generate incremental WAL records for pages
-                                // that didn't exist in the pageserver. Trying to replay those
-                                // WAL records failed to find the previous image of the page.
-                                // This special case allows us to recover from that situation.
-                                // See https://github.com/neondatabase/neon/issues/2601.
-                                //
-                                // Unfortunately we cannot do this for the main fork, or for
-                                // any metadata keys, keys, as that would lead to actual data
-                                // loss.
-                                if is_rel_fsm_block_key(img_key) || is_rel_vm_block_key(img_key) {
-                                    warn!("could not reconstruct FSM or VM key {img_key}, filling with zeros: {err:?}");
-                                    ZERO_PAGE.clone()
-                                } else {
-                                    return Err(CreateImageLayersError::PageReconstructError(err));
-                                }
-                            }
-                        };
-
-                        // Write all the keys we just read into our new image layer.
-                        image_layer_writer.put_image(img_key, img, ctx).await?;
-                        wrote_keys = true;
-                    }
-                }
-            }
-        }
-
-        if wrote_keys {
-            // Normal path: we have written some data into the new image layer for this
-            // partition, so flush it to disk.
-            let image_layer = image_layer_writer.finish(self, ctx).await?;
-            Ok(ImageLayerCreationOutcome {
-                image: Some(image_layer),
-                next_start_key: img_range.end,
-            })
-        } else {
-            // Special case: the image layer may be empty if this is a sharded tenant and the
-            // partition does not cover any keys owned by this shard.  In this case, to ensure
-            // we don't leave gaps between image layers, leave `start` where it is, so that the next
-            // layer we write will cover the key range that we just scanned.
-            tracing::debug!("no data in range {}-{}", img_range.start, img_range.end);
-            Ok(ImageLayerCreationOutcome {
-                image: None,
-                next_start_key: start,
-            })
-        }
-    }
-
-    /// Create an image layer for metadata keys. This function produces one image layer for all metadata
-    /// keys for now. Because metadata keys cannot exceed basebackup size limit, the image layer for it
-    /// would not be too large to fit in a single image layer.
-    #[allow(clippy::too_many_arguments)]
-    async fn create_image_layer_for_metadata_keys(
-        self: &Arc<Self>,
-        partition: &KeySpace,
-        mut image_layer_writer: ImageLayerWriter,
-        lsn: Lsn,
-        ctx: &RequestContext,
-        img_range: Range<Key>,
-        mode: ImageLayerCreationMode,
-    ) -> Result<ImageLayerCreationOutcome, CreateImageLayersError> {
-        assert!(!matches!(mode, ImageLayerCreationMode::Initial));
-
-        // Metadata keys image layer creation.
-        let mut reconstruct_state = ValuesReconstructState::default();
-        let data = self
-            .get_vectored_impl(partition.clone(), lsn, &mut reconstruct_state, ctx)
-            .await?;
-        let (data, total_kb_retrieved, total_key_retrieved) = {
-            let mut new_data = BTreeMap::new();
-            let mut total_kb_retrieved = 0;
-            let mut total_key_retrieved = 0;
-            for (k, v) in data {
-                let v = v.map_err(CreateImageLayersError::PageReconstructError)?;
-                total_kb_retrieved += KEY_SIZE + v.len();
-                total_key_retrieved += 1;
-                new_data.insert(k, v);
-            }
-            (new_data, total_kb_retrieved / 1024, total_key_retrieved)
-        };
-        let delta_file_accessed = reconstruct_state.get_delta_layers_visited();
-
-        let trigger_generation = delta_file_accessed as usize >= MAX_AUX_FILE_V2_DELTAS;
-        debug!(
-            "generate image layers for metadata keys: trigger_generation={trigger_generation}, \
-                delta_file_accessed={delta_file_accessed}, total_kb_retrieved={total_kb_retrieved}, \
-                total_key_retrieved={total_key_retrieved}"
-        );
-        if !trigger_generation && mode == ImageLayerCreationMode::Try {
-            return Ok(ImageLayerCreationOutcome {
-                image: None,
-                next_start_key: img_range.end,
-            });
-        }
-        let has_keys = !data.is_empty();
-        for (k, v) in data {
-            // Even if the value is empty (deleted), we do not delete it for now until we can ensure vectored get
-            // considers this situation properly.
-            // if v.is_empty() {
-            //     continue;
-            // }
-
-            // No need to handle sharding b/c metadata keys are always on the 0-th shard.
-
-            // TODO: split image layers to avoid too large layer files. Too large image files are not handled
-            // on the normal data path either.
-            image_layer_writer.put_image(k, v, ctx).await?;
-        }
-        Ok(ImageLayerCreationOutcome {
-            image: if has_keys {
-                let image_layer = image_layer_writer.finish(self, ctx).await?;
-                Some(image_layer)
-            } else {
-                tracing::debug!("no data in range {}-{}", img_range.start, img_range.end);
-                None
-            },
-            next_start_key: img_range.end,
-        })
-    }
-
     #[tracing::instrument(skip_all, fields(%lsn, %mode))]
     async fn create_image_layers(
         self: &Arc<Timeline>,
@@ -4413,17 +4168,19 @@ impl Timeline {
 
         for partition in partitioning.parts.iter() {
             let img_range = start..partition.ranges.last().unwrap().end;
-            let compact_metadata = partition.overlaps(&Key::metadata_key_range());
-            if compact_metadata {
-                for range in &partition.ranges {
-                    assert!(
-                        range.start.field1 >= METADATA_KEY_BEGIN_PREFIX
-                            && range.end.field1 <= METADATA_KEY_END_PREFIX,
-                        "metadata keys must be partitioned separately"
-                    );
-                }
-                if mode == ImageLayerCreationMode::Initial {
-                    return Err(CreateImageLayersError::Other(anyhow::anyhow!("no image layer should be created for metadata keys when flushing frozen layers")));
+
+            if partition.overlaps(&Key::metadata_key_range()) {
+                // TODO(chi): The next patch will correctly create image layers for metadata keys, and it would be a
+                // rather big change. Keep this patch small for now.
+                match mode {
+                    ImageLayerCreationMode::Force | ImageLayerCreationMode::Try => {
+                        // skip image layer creation anyways for metadata keys.
+                        start = img_range.end;
+                        continue;
+                    }
+                    ImageLayerCreationMode::Initial => {
+                        return Err(CreateImageLayersError::Other(anyhow::anyhow!("no image layer should be created for metadata keys when flushing frozen layers")));
+                    }
                 }
             } else if let ImageLayerCreationMode::Try = mode {
                 // check_for_image_layers = false -> skip
@@ -4434,7 +4191,7 @@ impl Timeline {
                 }
             }
 
-            let image_layer_writer = ImageLayerWriter::new(
+            let mut image_layer_writer = ImageLayerWriter::new(
                 self.conf,
                 self.timeline_id,
                 self.tenant_shard_id,
@@ -4450,39 +4207,87 @@ impl Timeline {
                 )))
             });
 
-            if !compact_metadata {
-                let ImageLayerCreationOutcome {
-                    image,
-                    next_start_key,
-                } = self
-                    .create_image_layer_for_rel_blocks(
-                        partition,
-                        image_layer_writer,
-                        lsn,
-                        ctx,
-                        img_range,
-                        start,
-                    )
-                    .await?;
+            let mut wrote_keys = false;
 
-                start = next_start_key;
-                image_layers.extend(image);
+            let mut key_request_accum = KeySpaceAccum::new();
+            for range in &partition.ranges {
+                let mut key = range.start;
+                while key < range.end {
+                    // Decide whether to retain this key: usually we do, but sharded tenants may
+                    // need to drop keys that don't belong to them.  If we retain the key, add it
+                    // to `key_request_accum` for later issuing a vectored get
+                    if self.shard_identity.is_key_disposable(&key) {
+                        debug!(
+                            "Dropping key {} during compaction (it belongs on shard {:?})",
+                            key,
+                            self.shard_identity.get_shard_number(&key)
+                        );
+                    } else {
+                        key_request_accum.add_key(key);
+                    }
+
+                    let last_key_in_range = key.next() == range.end;
+                    key = key.next();
+
+                    // Maybe flush `key_rest_accum`
+                    if key_request_accum.raw_size() >= Timeline::MAX_GET_VECTORED_KEYS
+                        || (last_key_in_range && key_request_accum.raw_size() > 0)
+                    {
+                        let results = self
+                            .get_vectored(key_request_accum.consume_keyspace(), lsn, ctx)
+                            .await?;
+
+                        for (img_key, img) in results {
+                            let img = match img {
+                                Ok(img) => img,
+                                Err(err) => {
+                                    // If we fail to reconstruct a VM or FSM page, we can zero the
+                                    // page without losing any actual user data. That seems better
+                                    // than failing repeatedly and getting stuck.
+                                    //
+                                    // We had a bug at one point, where we truncated the FSM and VM
+                                    // in the pageserver, but the Postgres didn't know about that
+                                    // and continued to generate incremental WAL records for pages
+                                    // that didn't exist in the pageserver. Trying to replay those
+                                    // WAL records failed to find the previous image of the page.
+                                    // This special case allows us to recover from that situation.
+                                    // See https://github.com/neondatabase/neon/issues/2601.
+                                    //
+                                    // Unfortunately we cannot do this for the main fork, or for
+                                    // any metadata keys, keys, as that would lead to actual data
+                                    // loss.
+                                    if is_rel_fsm_block_key(img_key) || is_rel_vm_block_key(img_key)
+                                    {
+                                        warn!("could not reconstruct FSM or VM key {img_key}, filling with zeros: {err:?}");
+                                        ZERO_PAGE.clone()
+                                    } else {
+                                        return Err(CreateImageLayersError::PageReconstructError(
+                                            err,
+                                        ));
+                                    }
+                                }
+                            };
+
+                            // Write all the keys we just read into our new image layer.
+                            image_layer_writer.put_image(img_key, img, ctx).await?;
+                            wrote_keys = true;
+                        }
+                    }
+                }
+            }
+
+            if wrote_keys {
+                // Normal path: we have written some data into the new image layer for this
+                // partition, so flush it to disk.
+                start = img_range.end;
+                let image_layer = image_layer_writer.finish(self, ctx).await?;
+                image_layers.push(image_layer);
             } else {
-                let ImageLayerCreationOutcome {
-                    image,
-                    next_start_key,
-                } = self
-                    .create_image_layer_for_metadata_keys(
-                        partition,
-                        image_layer_writer,
-                        lsn,
-                        ctx,
-                        img_range,
-                        mode,
-                    )
-                    .await?;
-                start = next_start_key;
-                image_layers.extend(image);
+                // Special case: the image layer may be empty if this is a sharded tenant and the
+                // partition does not cover any keys owned by this shard.  In this case, to ensure
+                // we don't leave gaps between image layers, leave `start` where it is, so that the next
+                // layer we write will cover the key range that we just scanned.
+                tracing::debug!("no data in range {}-{}", img_range.start, img_range.end);
             }
         }
 
@@ -4596,14 +4401,6 @@ impl Timeline {
     ) -> Result<Vec<TimelineId>, anyhow::Error> {
         detach_ancestor::complete(self, tenant, prepared, ctx).await
     }
-
-    /// Switch aux file policy and schedule upload to the index part.
-    pub(crate) fn do_switch_aux_policy(&self, policy: AuxFilePolicy) -> anyhow::Result<()> {
-        self.last_aux_file_policy.store(Some(policy));
-        self.remote_client
-            .schedule_index_upload_for_aux_file_policy_update(Some(policy))?;
-        Ok(())
-    }
 }
 
 /// Top-level failure to compact.
@@ -4860,32 +4657,7 @@ impl Timeline {
             (horizon_cutoff, pitr_cutoff, retain_lsns)
         };
 
-        let mut new_gc_cutoff = Lsn::min(horizon_cutoff, pitr_cutoff);
-        let standby_horizon = self.standby_horizon.load();
-        // Hold GC for the standby, but as a safety guard do it only within some
-        // reasonable lag.
-        if standby_horizon != Lsn::INVALID {
-            if let Some(standby_lag) = new_gc_cutoff.checked_sub(standby_horizon) {
-                const MAX_ALLOWED_STANDBY_LAG: u64 = 10u64 << 30; // 10 GB
-                if standby_lag.0 < MAX_ALLOWED_STANDBY_LAG {
-                    new_gc_cutoff = Lsn::min(standby_horizon, new_gc_cutoff);
-                    trace!("holding off GC for standby apply LSN {}", standby_horizon);
-                } else {
-                    warn!(
-                        "standby is lagging for more than {}MB, not holding gc for it",
-                        MAX_ALLOWED_STANDBY_LAG / 1024 / 1024
-                    )
-                }
-            }
-        }
-
-        // Reset standby horizon to ignore it if it is not updated till next GC.
-        // It is an easy way to unset it when standby disappears without adding
-        // more conf options.
-        self.standby_horizon.store(Lsn::INVALID);
-        self.metrics
-            .standby_horizon_gauge
-            .set(Lsn::INVALID.0 as i64);
+        let new_gc_cutoff = Lsn::min(horizon_cutoff, pitr_cutoff);
 
         let res = self
             .gc_timeline(horizon_cutoff, pitr_cutoff, retain_lsns, new_gc_cutoff)

pageserver/src/tenant/timeline/compaction.rs

@@ -9,10 +9,7 @@ use std::ops::{Deref, Range};
 use std::sync::Arc;
 
 use super::layer_manager::LayerManager;
-use super::{
-    CompactFlags, CreateImageLayersError, DurationRecorder, ImageLayerCreationMode,
-    RecordedDuration, Timeline,
-};
+use super::{CompactFlags, DurationRecorder, ImageLayerCreationMode, RecordedDuration, Timeline};
 
 use anyhow::{anyhow, Context};
 use enumset::EnumSet;
@@ -25,13 +22,14 @@ use tracing::{debug, info, info_span, trace, warn, Instrument};
 use utils::id::TimelineId;
 
 use crate::context::{AccessStatsBehavior, RequestContext, RequestContextBuilder};
-use crate::page_cache;
 use crate::tenant::storage_layer::{AsLayerDesc, PersistentLayerDesc};
-use crate::tenant::timeline::{drop_rlock, Hole, ImageLayerCreationOutcome};
+use crate::tenant::timeline::{drop_rlock, is_rel_fsm_block_key, is_rel_vm_block_key, Hole};
 use crate::tenant::timeline::{DeltaLayerWriter, ImageLayerWriter};
 use crate::tenant::timeline::{Layer, ResidentLayer};
 use crate::tenant::DeltaLayer;
+use crate::tenant::PageReconstructError;
 use crate::virtual_file::{MaybeFatalIo, VirtualFile};
+use crate::{page_cache, ZERO_PAGE};
 
 use crate::keyspace::KeySpace;
 use crate::repository::Key;
@@ -118,13 +116,9 @@ impl Timeline {
 
                 // 3. Create new image layers for partitions that have been modified
                 // "enough".
-                let mut partitioning = dense_partitioning;
-                partitioning
-                    .parts
-                    .extend(sparse_partitioning.into_dense().parts);
-                let image_layers = self
+                let dense_layers = self
                     .create_image_layers(
-                        &partitioning,
+                        &dense_partitioning,
                         lsn,
                         if flags.contains(CompactFlags::ForceImageLayerCreation) {
                             ImageLayerCreationMode::Force
@@ -136,8 +130,24 @@ impl Timeline {
                     .await
                     .map_err(anyhow::Error::from)?;
 
-                self.upload_new_image_layers(image_layers)?;
-                partitioning.parts.len()
+                // For now, nothing will be produced...
+                let sparse_layers = self
+                    .create_image_layers(
+                        &sparse_partitioning.clone().into_dense(),
+                        lsn,
+                        if flags.contains(CompactFlags::ForceImageLayerCreation) {
+                            ImageLayerCreationMode::Force
+                        } else {
+                            ImageLayerCreationMode::Try
+                        },
+                        &image_ctx,
+                    )
+                    .await
+                    .map_err(anyhow::Error::from)?;
+                assert!(sparse_layers.is_empty());
+
+                self.upload_new_image_layers(dense_layers)?;
+                dense_partitioning.parts.len()
             }
             Err(err) => {
                 // no partitioning? This is normal, if the timeline was just created
@@ -489,11 +499,8 @@ impl Timeline {
 
         for &DeltaEntry { key: next_key, .. } in all_keys.iter() {
             if let Some(prev_key) = prev {
-                // just first fast filter, do not create hole entries for metadata keys. The last hole in the
-                // compaction is the gap between data key and metadata keys.
-                if next_key.to_i128() - prev_key.to_i128() >= min_hole_range
-                    && !Key::is_metadata_key(&prev_key)
-                {
+                // just first fast filter
+                if next_key.to_i128() - prev_key.to_i128() >= min_hole_range {
                     let key_range = prev_key..next_key;
                     // Measuring hole by just subtraction of i128 representation of key range boundaries
                     // has not so much sense, because largest holes will corresponds field1/field2 changes.
@@ -1152,10 +1159,10 @@ impl TimelineAdaptor {
         lsn: Lsn,
         key_range: &Range<Key>,
         ctx: &RequestContext,
-    ) -> Result<(), CreateImageLayersError> {
+    ) -> Result<(), PageReconstructError> {
         let timer = self.timeline.metrics.create_images_time_histo.start_timer();
 
-        let image_layer_writer = ImageLayerWriter::new(
+        let mut image_layer_writer = ImageLayerWriter::new(
             self.timeline.conf,
             self.timeline.timeline_id,
             self.timeline.tenant_shard_id,
@@ -1166,34 +1173,47 @@ impl TimelineAdaptor {
         .await?;
 
         fail_point!("image-layer-writer-fail-before-finish", |_| {
-            Err(CreateImageLayersError::Other(anyhow::anyhow!(
+            Err(PageReconstructError::Other(anyhow::anyhow!(
                 "failpoint image-layer-writer-fail-before-finish"
             )))
         });
-
-        let keyspace = KeySpace {
-            ranges: self.get_keyspace(key_range, lsn, ctx).await?,
-        };
-        // TODO set proper (stateful) start. The create_image_layer_for_rel_blocks function mostly
-        let start = Key::MIN;
-        let ImageLayerCreationOutcome {
-            image,
-            next_start_key: _,
-        } = self
-            .timeline
-            .create_image_layer_for_rel_blocks(
-                &keyspace,
-                image_layer_writer,
-                lsn,
-                ctx,
-                key_range.clone(),
-                start,
-            )
-            .await?;
-
-        if let Some(image_layer) = image {
-            self.new_images.push(image_layer);
+        let keyspace_ranges = self.get_keyspace(key_range, lsn, ctx).await?;
+        for range in &keyspace_ranges {
+            let mut key = range.start;
+            while key < range.end {
+                let img = match self.timeline.get(key, lsn, ctx).await {
+                    Ok(img) => img,
+                    Err(err) => {
+                        // If we fail to reconstruct a VM or FSM page, we can zero the
+                        // page without losing any actual user data. That seems better
+                        // than failing repeatedly and getting stuck.
+                        //
+                        // We had a bug at one point, where we truncated the FSM and VM
+                        // in the pageserver, but the Postgres didn't know about that
+                        // and continued to generate incremental WAL records for pages
+                        // that didn't exist in the pageserver. Trying to replay those
+                        // WAL records failed to find the previous image of the page.
+                        // This special case allows us to recover from that situation.
+                        // See https://github.com/neondatabase/neon/issues/2601.
+                        //
+                        // Unfortunately we cannot do this for the main fork, or for
+                        // any metadata keys, keys, as that would lead to actual data
+                        // loss.
+                        if is_rel_fsm_block_key(key) || is_rel_vm_block_key(key) {
+                            warn!("could not reconstruct FSM or VM key {key}, filling with zeros: {err:?}");
+                            ZERO_PAGE.clone()
+                        } else {
+                            return Err(err);
+                        }
+                    }
+                };
+                image_layer_writer.put_image(key, img, ctx).await?;
+                key = key.next();
+            }
         }
+        let image_layer = image_layer_writer.finish(&self.timeline, ctx).await?;
+
+        self.new_images.push(image_layer);
 
         timer.stop_and_record();
 

pageserver/src/tenant/timeline/delete.rs

@@ -280,8 +280,6 @@ impl DeleteTimelineFlow {
                 // Important. We dont pass ancestor above because it can be missing.
                 // Thus we need to skip the validation here.
                 CreateTimelineCause::Delete,
-                // Aux file policy is not needed for deletion, assuming deletion does not read aux keyspace
-                None,
             )
             .context("create_timeline_struct")?;
 

pageserver/src/tenant/timeline/detach_ancestor.rs

@@ -12,7 +12,7 @@ use crate::{
 };
 use tokio_util::sync::CancellationToken;
 use tracing::Instrument;
-use utils::{completion, generation::Generation, http::error::ApiError, id::TimelineId, lsn::Lsn};
+use utils::{completion, generation::Generation, id::TimelineId, lsn::Lsn};
 
 #[derive(Debug, thiserror::Error)]
 pub(crate) enum Error {
@@ -41,27 +41,6 @@ pub(crate) enum Error {
     Unexpected(#[source] anyhow::Error),
 }
 
-impl From<Error> for ApiError {
-    fn from(value: Error) -> Self {
-        match value {
-            e @ Error::NoAncestor => ApiError::Conflict(e.to_string()),
-            // TODO: ApiError converts the anyhow using debug formatting ... just stop using ApiError?
-            e @ Error::TooManyAncestors => ApiError::BadRequest(anyhow::anyhow!("{}", e)),
-            Error::ShuttingDown => ApiError::ShuttingDown,
-            Error::OtherTimelineDetachOngoing(_) => {
-                ApiError::ResourceUnavailable("other timeline detach is already ongoing".into())
-            }
-            // All of these contain shutdown errors, in fact, it's the most common
-            e @ Error::FlushAncestor(_)
-            | e @ Error::RewrittenDeltaDownloadFailed(_)
-            | e @ Error::CopyDeltaPrefix(_)
-            | e @ Error::UploadRewritten(_)
-            | e @ Error::CopyFailed(_)
-            | e @ Error::Unexpected(_) => ApiError::InternalServerError(e.into()),
-        }
-    }
-}
-
 pub(crate) struct PreparedTimelineDetach {
     layers: Vec<Layer>,
 }
@@ -77,7 +56,7 @@ impl Default for Options {
     fn default() -> Self {
         Self {
             rewrite_concurrency: std::num::NonZeroUsize::new(2).unwrap(),
-            copy_concurrency: std::num::NonZeroUsize::new(100).unwrap(),
+            copy_concurrency: std::num::NonZeroUsize::new(10).unwrap(),
         }
     }
 }
@@ -96,11 +75,6 @@ pub(super) async fn prepare(
         .as_ref()
         .map(|tl| (tl.clone(), detached.ancestor_lsn))
     else {
-        // TODO: check if we have already been detached; for this we need to read the stored data
-        // on remote client, for that we need a follow-up which makes uploads cheaper and maintains
-        // a projection of the commited data.
-        //
-        // the error is wrong per openapi
         return Err(NoAncestor);
     };
 
@@ -110,7 +84,7 @@ pub(super) async fn prepare(
 
     if ancestor.ancestor_timeline.is_some() {
         // non-technical requirement; we could flatten N ancestors just as easily but we chose
-        // not to, at least initially
+        // not to
         return Err(TooManyAncestors);
     }
 

pageserver/src/tenant/timeline/layer_manager.rs

@@ -166,7 +166,7 @@ impl LayerManager {
     /// Flush a frozen layer and add the written delta layer to the layer map.
     pub(crate) fn finish_flush_l0_layer(
         &mut self,
-        delta_layers: &[ResidentLayer],
+        delta_layer: Option<&ResidentLayer>,
         frozen_layer_for_check: &Arc<InMemoryLayer>,
         metrics: &TimelineMetrics,
     ) {
@@ -181,12 +181,10 @@ impl LayerManager {
         // layer to disk at the same time, that would not work.
         assert_eq!(Arc::as_ptr(&inmem), Arc::as_ptr(frozen_layer_for_check));
 
-        if !delta_layers.is_empty() {
+        if let Some(l) = delta_layer {
             let mut updates = self.layer_map.batch_update();
-            for l in delta_layers {
-                Self::insert_historic_layer(l.as_ref().clone(), &mut updates, &mut self.layer_fmgr);
-                metrics.record_new_file_metrics(l.layer_desc().file_size);
-            }
+            Self::insert_historic_layer(l.as_ref().clone(), &mut updates, &mut self.layer_fmgr);
+            metrics.record_new_file_metrics(l.layer_desc().file_size);
             updates.flush();
         }
     }

pageserver/src/tenant/timeline/walreceiver/connection_manager.rs

@@ -705,7 +705,6 @@ impl ConnectionManagerState {
                     commit_lsn: info.commit_lsn,
                     safekeeper_connstr: info.safekeeper_connstr,
                     availability_zone: info.availability_zone,
-                    standby_horizon: info.standby_horizon,
                 }
             }
             MessageType::SafekeeperDiscoveryResponse => {
@@ -726,21 +725,6 @@ impl ConnectionManagerState {
 
         WALRECEIVER_BROKER_UPDATES.inc();
 
-        trace!(
-            "safekeeper info update: standby_horizon(cutoff)={}",
-            timeline_update.standby_horizon
-        );
-        if timeline_update.standby_horizon != 0 {
-            // ignore reports from safekeepers not connected to replicas
-            self.timeline
-                .standby_horizon
-                .store(Lsn(timeline_update.standby_horizon));
-            self.timeline
-                .metrics
-                .standby_horizon_gauge
-                .set(timeline_update.standby_horizon as i64);
-        }
-
         let new_safekeeper_id = NodeId(timeline_update.safekeeper_id);
         let old_entry = self.wal_stream_candidates.insert(
             new_safekeeper_id,
@@ -1110,7 +1094,6 @@ mod tests {
                 commit_lsn,
                 safekeeper_connstr: safekeeper_connstr.to_owned(),
                 availability_zone: None,
-                standby_horizon: 0,
             },
             latest_update,
         }

pageserver/src/tenant/upload_queue.rs

@@ -8,7 +8,6 @@ use std::collections::{HashMap, VecDeque};
 use std::fmt::Debug;
 
 use chrono::NaiveDateTime;
-use pageserver_api::models::AuxFilePolicy;
 use std::sync::Arc;
 use tracing::info;
 use utils::lsn::AtomicLsn;
@@ -61,9 +60,6 @@ pub(crate) struct UploadQueueInitialized {
     /// Part of the flattened "next" `index_part.json`.
     pub(crate) latest_lineage: Lineage,
 
-    /// The last aux file policy used on this timeline.
-    pub(crate) last_aux_file_policy: Option<AuxFilePolicy>,
-
     /// `disk_consistent_lsn` from the last metadata file that was successfully
     /// uploaded. `Lsn(0)` if nothing was uploaded yet.
     /// Unlike `latest_files` or `latest_metadata`, this value is never ahead.
@@ -193,7 +189,6 @@ impl UploadQueue {
             dangling_files: HashMap::new(),
             shutting_down: false,
             shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
-            last_aux_file_policy: Default::default(),
         };
 
         *self = UploadQueue::Initialized(state);
@@ -244,7 +239,6 @@ impl UploadQueue {
             dangling_files: HashMap::new(),
             shutting_down: false,
             shutdown_ready: Arc::new(tokio::sync::Semaphore::new(0)),
-            last_aux_file_policy: index_part.last_aux_file_policy(),
         };
 
         *self = UploadQueue::Initialized(state);

patches/pgvector.patch

@@ -1,78 +0,0 @@
-From 0b0194a57bd0f3598bd57dbedd0df3932330169d Mon Sep 17 00:00:00 2001
-From: Heikki Linnakangas <heikki.linnakangas@iki.fi>
-Date: Fri, 2 Feb 2024 22:26:45 +0200
-Subject: [PATCH 1/1] Make v0.6.0 work with Neon
-
-Now that the WAL-logging happens as a separate step at the end of the
-build, we need a few neon-specific hints to make it work.
----
- src/hnswbuild.c | 36 ++++++++++++++++++++++++++++++++++++
- 1 file changed, 36 insertions(+)
-
-diff --git a/src/hnswbuild.c b/src/hnswbuild.c
-index 680789b..ec54dea 100644
---- a/src/hnswbuild.c
-+++ b/src/hnswbuild.c
-@@ -840,9 +840,17 @@ HnswParallelBuildMain(dsm_segment *seg, shm_toc *toc)
- 
- 	hnswarea = shm_toc_lookup(toc, PARALLEL_KEY_HNSW_AREA, false);
- 
-+#ifdef NEON_SMGR
-+	smgr_start_unlogged_build(RelationGetSmgr(indexRel));
-+#endif
-+
- 	/* Perform inserts */
- 	HnswParallelScanAndInsert(heapRel, indexRel, hnswshared, hnswarea, false);
- 
-+#ifdef NEON_SMGR
-+	smgr_finish_unlogged_build_phase_1(RelationGetSmgr(indexRel));
-+#endif
-+
- 	/* Close relations within worker */
- 	index_close(indexRel, indexLockmode);
- 	table_close(heapRel, heapLockmode);
-@@ -1089,13 +1097,41 @@ BuildIndex(Relation heap, Relation index, IndexInfo *indexInfo,
- 	SeedRandom(42);
- #endif
- 
-+#ifdef NEON_SMGR
-+	smgr_start_unlogged_build(RelationGetSmgr(index));
-+#endif
-+
- 	InitBuildState(buildstate, heap, index, indexInfo, forkNum);
- 
- 	BuildGraph(buildstate, forkNum);
- 
-+#ifdef NEON_SMGR
-+	smgr_finish_unlogged_build_phase_1(RelationGetSmgr(index));
-+#endif
-+
- 	if (RelationNeedsWAL(index))
-+	{
- 		log_newpage_range(index, forkNum, 0, RelationGetNumberOfBlocks(index), true);
- 
-+#ifdef NEON_SMGR
-+		{
-+#if PG_VERSION_NUM >= 160000
-+			RelFileLocator rlocator = RelationGetSmgr(index)->smgr_rlocator.locator;
-+#else
-+			RelFileNode rlocator = RelationGetSmgr(index)->smgr_rnode.node;
-+#endif
-+
-+			SetLastWrittenLSNForBlockRange(XactLastRecEnd, rlocator,
-+										   MAIN_FORKNUM, 0, RelationGetNumberOfBlocks(index));
-+			SetLastWrittenLSNForRelation(XactLastRecEnd, rlocator, MAIN_FORKNUM);
-+		}
-+#endif
-+	}
-+
-+#ifdef NEON_SMGR
-+	smgr_end_unlogged_build(RelationGetSmgr(index));
-+#endif
-+
- 	FreeBuildState(buildstate);
- }
- 
--- 
-2.39.2
-

pgxn/neon/libpagestore.c

@@ -49,7 +49,7 @@ char	   *neon_auth_token;
 int			readahead_buffer_size = 128;
 int			flush_every_n_requests = 8;
 
-int         neon_protocol_version = 2;
+int         neon_protocol_version = 1;
 
 static int	n_reconnect_attempts = 0;
 static int	max_reconnect_attempts = 60;
@@ -860,7 +860,7 @@ pg_init_libpagestore(void)
 							"Version of compute<->page server protocol",
 							NULL,
 							&neon_protocol_version,
-							2, /* use protocol version 2 */
+							1, /* default to old protocol for now */
 							1, /* min */
 							2, /* max */
 							PGC_SU_BACKEND,

pgxn/neon/pagestore_smgr.c

@@ -45,7 +45,6 @@
  */
 #include "postgres.h"
 
-#include "access/parallel.h"
 #include "access/xact.h"
 #include "access/xlog.h"
 #include "access/xlogdefs.h"
@@ -1349,10 +1348,6 @@ PageIsEmptyHeapPage(char *buffer)
 	return memcmp(buffer, empty_page.data, BLCKSZ) == 0;
 }
 
-/*
- * A page is being evicted from the shared buffer cache. Update the
- * last-written LSN of the page, and WAL-log it if needed.
- */
 static void
 #if PG_MAJORVERSION_NUM < 16
 neon_wallog_page(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, char *buffer, bool force)
@@ -1361,7 +1356,12 @@ neon_wallog_page(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, co
 #endif
 {
 	XLogRecPtr	lsn = PageGetLSN((Page) buffer);
-	bool		log_page;
+
+	if (ShutdownRequestPending)
+		return;
+	/* Don't log any pages if we're not allowed to do so. */
+	if (!XLogInsertAllowed())
+		return;
 
 	/*
 	 * Whenever a VM or FSM page is evicted, WAL-log it. FSM and (some) VM
@@ -1370,21 +1370,9 @@ neon_wallog_page(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, co
 	 * correctness, the non-logged updates are not critical. But we want to
 	 * have a reasonably up-to-date VM and FSM in the page server.
 	 */
-	log_page = false;
-	if (force)
-	{
-		Assert(XLogInsertAllowed());
-		log_page = true;
-	}
-	else if (XLogInsertAllowed() &&
-			 !ShutdownRequestPending &&
-			 (forknum == FSM_FORKNUM || forknum == VISIBILITYMAP_FORKNUM))
-	{
-		log_page = true;
-	}
-
-	if (log_page)
+	if ((force || forknum == FSM_FORKNUM || forknum == VISIBILITYMAP_FORKNUM) && !RecoveryInProgress())
 	{
+		/* FSM is never WAL-logged and we don't care. */
 		XLogRecPtr	recptr;
 
 		recptr = log_newpage_copy(&InfoFromSMgrRel(reln), forknum, blocknum,
@@ -1397,8 +1385,7 @@ neon_wallog_page(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, co
 						RelFileInfoFmt(InfoFromSMgrRel(reln)),
 						forknum, LSN_FORMAT_ARGS(lsn))));
 	}
-
-	if (lsn == InvalidXLogRecPtr)
+	else if (lsn == InvalidXLogRecPtr)
 	{
 		/*
 		 * When PostgreSQL extends a relation, it calls smgrextend() with an
@@ -1434,31 +1421,19 @@ neon_wallog_page(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum, co
 							RelFileInfoFmt(InfoFromSMgrRel(reln)),
 							forknum)));
 		}
-		else if (forknum != FSM_FORKNUM && forknum != VISIBILITYMAP_FORKNUM)
+		else
 		{
-			/*
-			 * Its a bad sign if there is a page with zero LSN in the buffer
-			 * cache in a standby, too. However, PANICing seems like a cure
-			 * worse than the disease, as the damage has likely already been
-			 * done in the primary. So in a standby, make this an assertion,
-			 * and in a release build just LOG the error and soldier on. We
-			 * update the last-written LSN of the page with a conservative
-			 * value in that case, which is the last replayed LSN.
-			 */
-			ereport(RecoveryInProgress() ? LOG : PANIC,
+			ereport(PANIC,
 					(errmsg(NEON_TAG "Page %u of relation %u/%u/%u.%u is evicted with zero LSN",
 							blocknum,
 							RelFileInfoFmt(InfoFromSMgrRel(reln)),
 							forknum)));
-			Assert(false);
-
-			lsn = GetXLogReplayRecPtr(NULL); /* in standby mode, soldier on */
 		}
 	}
 	else
 	{
 		ereport(SmgrTrace,
-				(errmsg(NEON_TAG "Evicting page %u of relation %u/%u/%u.%u with lsn=%X/%X",
+				(errmsg(NEON_TAG "Page %u of relation %u/%u/%u.%u is already wal logged at lsn=%X/%X",
 						blocknum,
 						RelFileInfoFmt(InfoFromSMgrRel(reln)),
 						forknum, LSN_FORMAT_ARGS(lsn))));
@@ -1551,92 +1526,8 @@ neon_get_request_lsns(NRelFileInfo rinfo, ForkNumber forknum, BlockNumber blkno)
 
 	if (RecoveryInProgress())
 	{
-		/*---
-		 * In broad strokes, a replica always requests the page at the current
-		 * replay LSN. But looking closer, what exactly is the replay LSN? Is
-		 * it the last replayed record, or the record being replayed? And does
-		 * the startup process performing the replay need to do something
-		 * differently than backends running queries? Let's take a closer look
-		 * at the different scenarios:
-		 *
-		 * 1. Startup process reads a page, last_written_lsn is old.
-		 *
-		 * Read the old version of the page. We will apply the WAL record on
-		 * it to bring it up-to-date.
-		 *
-		 * We could read the new version, with the changes from this WAL
-		 * record already applied, to offload the work of replaying the record
-		 * to the pageserver. The pageserver might not have received the WAL
-		 * record yet, though, so a read of the old page version and applying
-		 * the record ourselves is likely faster. Also, the redo function
-		 * might be surprised if the changes have already applied. That's
-		 * normal during crash recovery, but not in hot standby.
-		 *
-		 * 2. Startup process reads a page, last_written_lsn == record we're
-		 *    replaying.
-		 *
-		 * Can this happen? There are a few theoretical cases when it might:
-		 *
-		 * A) The redo function reads the same page twice. We had already read
-		 *    and applied the changes once, and now we're reading it for the
-		 *    second time.  That would be a rather silly thing for a redo
-		 *    function to do, and I'm not aware of any that would do it.
-		 *
-		 * B) The redo function modifies multiple pages, and it already
-		 *    applied the changes to one of the pages, released the lock on
-		 *    it, and is now reading a second page.  Furthermore, the first
-		 *    page was already evicted from the buffer cache, and also from
-		 *    the last-written LSN cache, so that the per-relation or global
-		 *    last-written LSN was already updated. All the WAL redo functions
-		 *    hold the locks on pages that they modify, until all the changes
-		 *    have been modified (?), which would make that impossible.
-		 *    However, we skip the locking, if the page isn't currently in the
-		 *    page cache (see neon_redo_read_buffer_filter below).
-		 *
-		 * Even if the one of the above cases were possible in theory, they
-		 * would also require the pages being modified by the redo function to
-		 * be immediately evicted from the page cache.
-		 *
-		 * So this probably does not happen in practice. But if it does, we
-		 * request the new version, including the changes from the record
-		 * being replayed. That seems like the correct behavior in any case.
-		 *
-		 * 3. Backend process reads a page with old last-written LSN
-		 *
-		 * Nothing special here. Read the old version.
-		 *
-		 * 4. Backend process reads a page with last_written_lsn == record being replayed
-		 *
-		 * This can happen, if the redo function has started to run, and saw
-		 * that the page isn't present in the page cache (see
-		 * neon_redo_read_buffer_filter below).  Normally, in a normal
-		 * Postgres server, the redo function would hold a lock on the page,
-		 * so we would get blocked waiting the redo function to release the
-		 * lock. To emulate that, wait for the WAL replay of the record to
-		 * finish.
-		 */
-		/* Request the page at the end of the last fully replayed LSN. */
-		XLogRecPtr replay_lsn = GetXLogReplayRecPtr(NULL);
-
-		if (last_written_lsn > replay_lsn)
-		{
-			/* GetCurrentReplayRecPtr was introduced in v15 */
-#if PG_VERSION_NUM >= 150000
-			Assert(last_written_lsn == GetCurrentReplayRecPtr(NULL));
-#endif
-
-			/*
-			 * Cases 2 and 4. If this is a backend (case 4), the
-			 * neon_read_at_lsn() call later will wait for the WAL record to be
-			 * fully replayed.
-			 */
-			result.request_lsn = last_written_lsn;
-		}
-		else
-		{
-			/* cases 1 and 3 */
-			result.request_lsn = replay_lsn;
-		}
+		/* Request the page at the last replayed LSN. */
+		result.request_lsn = GetXLogReplayRecPtr(NULL);
 		result.not_modified_since = last_written_lsn;
 		result.effective_request_lsn = result.request_lsn;
 		Assert(last_written_lsn <= result.request_lsn);
@@ -2931,14 +2822,10 @@ neon_start_unlogged_build(SMgrRelation reln)
 	reln->smgr_relpersistence = RELPERSISTENCE_UNLOGGED;
 
 	/*
-	 * Create the local file. In a parallel build, the leader is expected to
-	 * call this first and do it.
-	 *
 	 * FIXME: should we pass isRedo true to create the tablespace dir if it
 	 * doesn't exist? Is it needed?
 	 */
-	if (!IsParallelWorker())
-		mdcreate(reln, MAIN_FORKNUM, false);
+	mdcreate(reln, MAIN_FORKNUM, false);
 }
 
 /*
@@ -2962,17 +2849,7 @@ neon_finish_unlogged_build_phase_1(SMgrRelation reln)
 	Assert(unlogged_build_phase == UNLOGGED_BUILD_PHASE_1);
 	Assert(reln->smgr_relpersistence == RELPERSISTENCE_UNLOGGED);
 
-	/*
-	 * In a parallel build, (only) the leader process performs the 2nd
-	 * phase.
-	 */
-	if (IsParallelWorker())
-	{
-		unlogged_build_rel = NULL;
-		unlogged_build_phase = UNLOGGED_BUILD_NOT_IN_PROGRESS;
-	}
-	else
-		unlogged_build_phase = UNLOGGED_BUILD_PHASE_2;
+	unlogged_build_phase = UNLOGGED_BUILD_PHASE_2;
 }
 
 /*
@@ -3324,7 +3201,7 @@ neon_redo_read_buffer_filter(XLogReaderState *record, uint8 block_id)
 	BufferTag	tag;
 	uint32		hash;
 	LWLock	   *partitionLock;
-	int			buf_id;
+	Buffer		buffer;
 	bool		no_redo_needed;
 
 	if (old_redo_read_buffer_filter && old_redo_read_buffer_filter(record, block_id))
@@ -3362,20 +3239,20 @@ neon_redo_read_buffer_filter(XLogReaderState *record, uint8 block_id)
 	else
 	{
 		/* Try to find the relevant buffer */
-		buf_id = BufTableLookup(&tag, hash);
+		buffer = BufTableLookup(&tag, hash);
 
-		no_redo_needed = buf_id < 0;
+		no_redo_needed = buffer < 0;
 	}
+	/* In both cases st lwlsn past this WAL record */
+	SetLastWrittenLSNForBlock(end_recptr, rinfo, forknum, blkno);
 
 	/*
 	 * we don't have the buffer in memory, update lwLsn past this record, also
 	 * evict page from file cache
 	 */
 	if (no_redo_needed)
-	{
-		SetLastWrittenLSNForBlock(end_recptr, rinfo, forknum, blkno);
 		lfc_evict(rinfo, forknum, blkno);
-	}
+
 
 	LWLockRelease(partitionLock);
 

pgxn/neon/walproposer_pg.c

@@ -1852,30 +1852,34 @@ static void
 CombineHotStanbyFeedbacks(HotStandbyFeedback *hs, WalProposer *wp)
 {
 	hs->ts = 0;
-	hs->xmin = InvalidFullTransactionId;
-	hs->catalog_xmin = InvalidFullTransactionId;
+	hs->xmin.value = ~0;		/* largest unsigned value */
+	hs->catalog_xmin.value = ~0;	/* largest unsigned value */
 
 	for (int i = 0; i < wp->n_safekeepers; i++)
 	{
-
-		if (wp->safekeeper[i].state == SS_ACTIVE)
+		if (wp->safekeeper[i].appendResponse.hs.ts != 0)
 		{
 			HotStandbyFeedback *skhs = &wp->safekeeper[i].appendResponse.hs;
 
 			if (FullTransactionIdIsNormal(skhs->xmin)
-				&& (!FullTransactionIdIsValid(hs->xmin) || FullTransactionIdPrecedes(skhs->xmin, hs->xmin)))
+				&& FullTransactionIdPrecedes(skhs->xmin, hs->xmin))
 			{
 				hs->xmin = skhs->xmin;
 				hs->ts = skhs->ts;
 			}
 			if (FullTransactionIdIsNormal(skhs->catalog_xmin)
-				&& (!FullTransactionIdIsValid(hs->catalog_xmin) || FullTransactionIdPrecedes(skhs->catalog_xmin, hs->catalog_xmin)))
+				&& FullTransactionIdPrecedes(skhs->catalog_xmin, hs->xmin))
 			{
 				hs->catalog_xmin = skhs->catalog_xmin;
 				hs->ts = skhs->ts;
 			}
 		}
 	}
+
+	if (hs->xmin.value == ~0)
+		hs->xmin = InvalidFullTransactionId;
+	if (hs->catalog_xmin.value == ~0)
+		hs->catalog_xmin = InvalidFullTransactionId;
 }
 
 /*
@@ -1942,28 +1946,14 @@ walprop_pg_process_safekeeper_feedback(WalProposer *wp, Safekeeper *sk)
 	}
 
 	CombineHotStanbyFeedbacks(&hsFeedback, wp);
-	if (memcmp(&hsFeedback, &agg_hs_feedback, sizeof hsFeedback) != 0)
+	if (hsFeedback.ts != 0 && memcmp(&hsFeedback, &agg_hs_feedback, sizeof hsFeedback) != 0)
 	{
-		FullTransactionId xmin = hsFeedback.xmin;
-		FullTransactionId catalog_xmin = hsFeedback.catalog_xmin;
-		FullTransactionId next_xid = ReadNextFullTransactionId();
-		/*
-		 * Page server is updating nextXid in checkpoint each 1024 transactions,
-		 * so feedback xmin can be actually larger then nextXid and
-		 * function TransactionIdInRecentPast return false in this case,
-		 * preventing update of slot's xmin.
-		 */
-		if (FullTransactionIdPrecedes(next_xid, xmin))
-			xmin = next_xid;
-		if (FullTransactionIdPrecedes(next_xid, catalog_xmin))
-			catalog_xmin = next_xid;
 		agg_hs_feedback = hsFeedback;
-		elog(DEBUG2, "ProcessStandbyHSFeedback(xmin=%d, catalog_xmin=%d", XidFromFullTransactionId(hsFeedback.xmin), XidFromFullTransactionId(hsFeedback.catalog_xmin));
 		ProcessStandbyHSFeedback(hsFeedback.ts,
-								 XidFromFullTransactionId(xmin),
-								 EpochFromFullTransactionId(xmin),
-								 XidFromFullTransactionId(catalog_xmin),
-								 EpochFromFullTransactionId(catalog_xmin));
+								 XidFromFullTransactionId(hsFeedback.xmin),
+								 EpochFromFullTransactionId(hsFeedback.xmin),
+								 XidFromFullTransactionId(hsFeedback.catalog_xmin),
+								 EpochFromFullTransactionId(hsFeedback.catalog_xmin));
 	}
 
 	CheckGracefulShutdown(wp);

poetry.lock

@@ -2405,7 +2405,6 @@ files = [
     {file = "PyYAML-6.0.1-cp311-cp311-win_amd64.whl", hash = "sha256:bf07ee2fef7014951eeb99f56f39c9bb4af143d8aa3c21b1677805985307da34"},
     {file = "PyYAML-6.0.1-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:855fb52b0dc35af121542a76b9a84f8d1cd886ea97c84703eaa6d88e37a2ad28"},
     {file = "PyYAML-6.0.1-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:40df9b996c2b73138957fe23a16a4f0ba614f4c0efce1e9406a184b6d07fa3a9"},
-    {file = "PyYAML-6.0.1-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:a08c6f0fe150303c1c6b71ebcd7213c2858041a7e01975da3a99aed1e7a378ef"},
     {file = "PyYAML-6.0.1-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:6c22bec3fbe2524cde73d7ada88f6566758a8f7227bfbf93a408a9d86bcc12a0"},
     {file = "PyYAML-6.0.1-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:8d4e9c88387b0f5c7d5f281e55304de64cf7f9c0021a3525bd3b1c542da3b0e4"},
     {file = "PyYAML-6.0.1-cp312-cp312-win32.whl", hash = "sha256:d483d2cdf104e7c9fa60c544d92981f12ad66a457afae824d146093b8c294c54"},
@@ -2530,13 +2529,13 @@ files = [
 
 [[package]]
 name = "requests"
-version = "2.32.0"
+version = "2.31.0"
 description = "Python HTTP for Humans."
 optional = false
-python-versions = ">=3.8"
+python-versions = ">=3.7"
 files = [
-    {file = "requests-2.32.0-py3-none-any.whl", hash = "sha256:f2c3881dddb70d056c5bd7600a4fae312b2a300e39be6a118d30b90bd27262b5"},
-    {file = "requests-2.32.0.tar.gz", hash = "sha256:fa5490319474c82ef1d2c9bc459d3652e3ae4ef4c4ebdd18a21145a47ca4b6b8"},
+    {file = "requests-2.31.0-py3-none-any.whl", hash = "sha256:58cd2187c01e70e6e26505bca751777aa9f2ee0b7f4300988b709f44e013003f"},
+    {file = "requests-2.31.0.tar.gz", hash = "sha256:942c5a758f98d790eaed1a29cb6eefc7ffb0d1cf7af05c3d2791656dbd6ad1e1"},
 ]
 
 [package.dependencies]
@@ -2960,16 +2959,6 @@ files = [
     {file = "wrapt-1.14.1-cp310-cp310-musllinux_1_1_x86_64.whl", hash = "sha256:8ad85f7f4e20964db4daadcab70b47ab05c7c1cf2a7c1e51087bfaa83831854c"},
     {file = "wrapt-1.14.1-cp310-cp310-win32.whl", hash = "sha256:a9a52172be0b5aae932bef82a79ec0a0ce87288c7d132946d645eba03f0ad8a8"},
     {file = "wrapt-1.14.1-cp310-cp310-win_amd64.whl", hash = "sha256:6d323e1554b3d22cfc03cd3243b5bb815a51f5249fdcbb86fda4bf62bab9e164"},
-    {file = "wrapt-1.14.1-cp311-cp311-macosx_10_9_x86_64.whl", hash = "sha256:ecee4132c6cd2ce5308e21672015ddfed1ff975ad0ac8d27168ea82e71413f55"},
-    {file = "wrapt-1.14.1-cp311-cp311-macosx_11_0_arm64.whl", hash = "sha256:2020f391008ef874c6d9e208b24f28e31bcb85ccff4f335f15a3251d222b92d9"},
-    {file = "wrapt-1.14.1-cp311-cp311-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:2feecf86e1f7a86517cab34ae6c2f081fd2d0dac860cb0c0ded96d799d20b335"},
-    {file = "wrapt-1.14.1-cp311-cp311-manylinux_2_5_i686.manylinux1_i686.manylinux_2_17_i686.manylinux2014_i686.whl", hash = "sha256:240b1686f38ae665d1b15475966fe0472f78e71b1b4903c143a842659c8e4cb9"},
-    {file = "wrapt-1.14.1-cp311-cp311-manylinux_2_5_x86_64.manylinux1_x86_64.manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:a9008dad07d71f68487c91e96579c8567c98ca4c3881b9b113bc7b33e9fd78b8"},
-    {file = "wrapt-1.14.1-cp311-cp311-musllinux_1_1_aarch64.whl", hash = "sha256:6447e9f3ba72f8e2b985a1da758767698efa72723d5b59accefd716e9e8272bf"},
-    {file = "wrapt-1.14.1-cp311-cp311-musllinux_1_1_i686.whl", hash = "sha256:acae32e13a4153809db37405f5eba5bac5fbe2e2ba61ab227926a22901051c0a"},
-    {file = "wrapt-1.14.1-cp311-cp311-musllinux_1_1_x86_64.whl", hash = "sha256:49ef582b7a1152ae2766557f0550a9fcbf7bbd76f43fbdc94dd3bf07cc7168be"},
-    {file = "wrapt-1.14.1-cp311-cp311-win32.whl", hash = "sha256:358fe87cc899c6bb0ddc185bf3dbfa4ba646f05b1b0b9b5a27c2cb92c2cea204"},
-    {file = "wrapt-1.14.1-cp311-cp311-win_amd64.whl", hash = "sha256:26046cd03936ae745a502abf44dac702a5e6880b2b01c29aea8ddf3353b68224"},
     {file = "wrapt-1.14.1-cp35-cp35m-manylinux1_i686.whl", hash = "sha256:43ca3bbbe97af00f49efb06e352eae40434ca9d915906f77def219b88e85d907"},
     {file = "wrapt-1.14.1-cp35-cp35m-manylinux1_x86_64.whl", hash = "sha256:6b1a564e6cb69922c7fe3a678b9f9a3c54e72b469875aa8018f18b4d1dd1adf3"},
     {file = "wrapt-1.14.1-cp35-cp35m-manylinux2010_i686.whl", hash = "sha256:00b6d4ea20a906c0ca56d84f93065b398ab74b927a7a3dbd470f6fc503f95dc3"},
@@ -3207,4 +3196,4 @@ cffi = ["cffi (>=1.11)"]
 [metadata]
 lock-version = "2.0"
 python-versions = "^3.9"
-content-hash = "16ebd6a46768be7f67dbdb4ee5903b167d94edc9965f29252f038c67e9e907b0"
+content-hash = "dcde14c58a32bda5f123319a069352c458b3719f3c62977991eebb9803a46a9e"

proxy/Cargo.toml

@@ -9,7 +9,6 @@ default = []
 testing = []
 
 [dependencies]
-ahash.workspace = true
 anyhow.workspace = true
 async-compression.workspace = true
 async-trait.workspace = true
@@ -25,10 +24,8 @@ camino.workspace = true
 chrono.workspace = true
 clap.workspace = true
 consumption_metrics.workspace = true
-crossbeam-deque.workspace = true
 dashmap.workspace = true
 env_logger.workspace = true
-framed-websockets.workspace = true
 futures.workspace = true
 git-version.workspace = true
 hashbrown.workspace = true
@@ -38,6 +35,7 @@ hmac.workspace = true
 hostname.workspace = true
 http.workspace = true
 humantime.workspace = true
+hyper-tungstenite.workspace = true
 hyper.workspace = true
 hyper1 = { package = "hyper", version = "1.2", features = ["server"] }
 hyper-util = { version = "0.1", features = ["server", "http1", "http2", "tokio"] }
@@ -54,6 +52,7 @@ opentelemetry.workspace = true
 parking_lot.workspace = true
 parquet.workspace = true
 parquet_derive.workspace = true
+pbkdf2 = { workspace = true, features = ["simple", "std"] }
 pin-project-lite.workspace = true
 postgres_backend.workspace = true
 pq_proto.workspace = true
@@ -77,6 +76,7 @@ smol_str.workspace = true
 smallvec.workspace = true
 socket2.workspace = true
 subtle.workspace = true
+sync_wrapper.workspace = true
 task-local-extensions.workspace = true
 thiserror.workspace = true
 tikv-jemallocator.workspace = true
@@ -106,8 +106,6 @@ workspace_hack.workspace = true
 [dev-dependencies]
 camino-tempfile.workspace = true
 fallible-iterator.workspace = true
-tokio-tungstenite.workspace = true
-pbkdf2 = { workspace = true, features = ["simple", "std"] }
 rcgen.workspace = true
 rstest.workspace = true
 tokio-postgres-rustls.workspace = true

proxy/src/auth/backend.rs

@@ -365,10 +365,7 @@ async fn authenticate_with_secret(
     config: &'static AuthenticationConfig,
 ) -> auth::Result<ComputeCredentials> {
     if let Some(password) = unauthenticated_password {
-        let ep = EndpointIdInt::from(&info.endpoint);
-
-        let auth_outcome =
-            validate_password_and_exchange(&config.thread_pool, ep, &password, secret).await?;
+        let auth_outcome = validate_password_and_exchange(&password, secret).await?;
         let keys = match auth_outcome {
             crate::sasl::Outcome::Success(key) => key,
             crate::sasl::Outcome::Failure(reason) => {
@@ -389,7 +386,7 @@ async fn authenticate_with_secret(
     // Currently, we use it for websocket connections (latency).
     if allow_cleartext {
         ctx.set_auth_method(crate::context::AuthMethod::Cleartext);
-        return hacks::authenticate_cleartext(ctx, info, client, secret, config).await;
+        return hacks::authenticate_cleartext(ctx, info, client, secret).await;
     }
 
     // Finally, proceed with the main auth flow (SCRAM-based).
@@ -557,7 +554,7 @@ mod tests {
         context::RequestMonitoring,
         proxy::NeonOptions,
         rate_limiter::{EndpointRateLimiter, RateBucketInfo},
-        scram::{threadpool::ThreadPool, ServerSecret},
+        scram::ServerSecret,
         stream::{PqStream, Stream},
     };
 
@@ -599,7 +596,6 @@ mod tests {
     }
 
     static CONFIG: Lazy<AuthenticationConfig> = Lazy::new(|| AuthenticationConfig {
-        thread_pool: ThreadPool::new(1),
         scram_protocol_timeout: std::time::Duration::from_secs(5),
         rate_limiter_enabled: true,
         rate_limiter: AuthRateLimiter::new(&RateBucketInfo::DEFAULT_AUTH_SET),

proxy/src/auth/backend/hacks.rs

@@ -3,10 +3,8 @@ use super::{
 };
 use crate::{
     auth::{self, AuthFlow},
-    config::AuthenticationConfig,
     console::AuthSecret,
     context::RequestMonitoring,
-    intern::EndpointIdInt,
     sasl,
     stream::{self, Stream},
 };
@@ -22,7 +20,6 @@ pub async fn authenticate_cleartext(
     info: ComputeUserInfo,
     client: &mut stream::PqStream<Stream<impl AsyncRead + AsyncWrite + Unpin>>,
     secret: AuthSecret,
-    config: &'static AuthenticationConfig,
 ) -> auth::Result<ComputeCredentials> {
     warn!("cleartext auth flow override is enabled, proceeding");
     ctx.set_auth_method(crate::context::AuthMethod::Cleartext);
@@ -30,14 +27,8 @@ pub async fn authenticate_cleartext(
     // pause the timer while we communicate with the client
     let paused = ctx.latency_timer.pause(crate::metrics::Waiting::Client);
 
-    let ep = EndpointIdInt::from(&info.endpoint);
-
     let auth_flow = AuthFlow::new(client)
-        .begin(auth::CleartextPassword {
-            secret,
-            endpoint: ep,
-            pool: config.thread_pool.clone(),
-        })
+        .begin(auth::CleartextPassword(secret))
         .await?;
     drop(paused);
     // cleartext auth is only allowed to the ws/http protocol.

proxy/src/auth/flow.rs

@@ -5,14 +5,12 @@ use crate::{
     config::TlsServerEndPoint,
     console::AuthSecret,
     context::RequestMonitoring,
-    intern::EndpointIdInt,
-    sasl,
-    scram::{self, threadpool::ThreadPool},
+    sasl, scram,
     stream::{PqStream, Stream},
 };
 use postgres_protocol::authentication::sasl::{SCRAM_SHA_256, SCRAM_SHA_256_PLUS};
 use pq_proto::{BeAuthenticationSaslMessage, BeMessage, BeMessage as Be};
-use std::{io, sync::Arc};
+use std::io;
 use tokio::io::{AsyncRead, AsyncWrite};
 use tracing::info;
 
@@ -55,11 +53,7 @@ impl AuthMethod for PasswordHack {
 
 /// Use clear-text password auth called `password` in docs
 /// <https://www.postgresql.org/docs/current/auth-password.html>
-pub struct CleartextPassword {
-    pub pool: Arc<ThreadPool>,
-    pub endpoint: EndpointIdInt,
-    pub secret: AuthSecret,
-}
+pub struct CleartextPassword(pub AuthSecret);
 
 impl AuthMethod for CleartextPassword {
     #[inline(always)]
@@ -132,13 +126,7 @@ impl<S: AsyncRead + AsyncWrite + Unpin> AuthFlow<'_, S, CleartextPassword> {
             .strip_suffix(&[0])
             .ok_or(AuthErrorImpl::MalformedPassword("missing terminator"))?;
 
-        let outcome = validate_password_and_exchange(
-            &self.state.pool,
-            self.state.endpoint,
-            password,
-            self.state.secret,
-        )
-        .await?;
+        let outcome = validate_password_and_exchange(password, self.state.0).await?;
 
         if let sasl::Outcome::Success(_) = &outcome {
             self.stream.write_message_noflush(&Be::AuthenticationOk)?;
@@ -193,8 +181,6 @@ impl<S: AsyncRead + AsyncWrite + Unpin> AuthFlow<'_, S, Scram<'_>> {
 }
 
 pub(crate) async fn validate_password_and_exchange(
-    pool: &ThreadPool,
-    endpoint: EndpointIdInt,
     password: &[u8],
     secret: AuthSecret,
 ) -> super::Result<sasl::Outcome<ComputeCredentialKeys>> {
@@ -208,7 +194,7 @@ pub(crate) async fn validate_password_and_exchange(
         }
         // perform scram authentication as both client and server to validate the keys
         AuthSecret::Scram(scram_secret) => {
-            let outcome = crate::scram::exchange(pool, endpoint, &scram_secret, password).await?;
+            let outcome = crate::scram::exchange(&scram_secret, password).await?;
 
             let client_key = match outcome {
                 sasl::Outcome::Success(client_key) => client_key,

proxy/src/bin/proxy.rs

@@ -27,7 +27,6 @@ use proxy::redis::cancellation_publisher::RedisPublisherClient;
 use proxy::redis::connection_with_credentials_provider::ConnectionWithCredentialsProvider;
 use proxy::redis::elasticache;
 use proxy::redis::notifications;
-use proxy::scram::threadpool::ThreadPool;
 use proxy::serverless::cancel_set::CancelSet;
 use proxy::serverless::GlobalConnPoolOptions;
 use proxy::usage_metrics;
@@ -133,9 +132,6 @@ struct ProxyCliArgs {
     /// timeout for scram authentication protocol
     #[clap(long, default_value = "15s", value_parser = humantime::parse_duration)]
     scram_protocol_timeout: tokio::time::Duration,
-    /// size of the threadpool for password hashing
-    #[clap(long, default_value_t = 4)]
-    scram_thread_pool_size: u8,
     /// Require that all incoming requests have a Proxy Protocol V2 packet **and** have an IP address associated.
     #[clap(long, default_value_t = false, value_parser = clap::builder::BoolishValueParser::new(), action = clap::ArgAction::Set)]
     require_client_ip: bool,
@@ -493,9 +489,6 @@ async fn main() -> anyhow::Result<()> {
 
 /// ProxyConfig is created at proxy startup, and lives forever.
 fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
-    let thread_pool = ThreadPool::new(args.scram_thread_pool_size);
-    Metrics::install(thread_pool.metrics.clone());
-
     let tls_config = match (&args.tls_key, &args.tls_cert) {
         (Some(key_path), Some(cert_path)) => Some(config::configure_tls(
             key_path,
@@ -631,7 +624,6 @@ fn build_config(args: &ProxyCliArgs) -> anyhow::Result<&'static ProxyConfig> {
         client_conn_threshold: args.sql_over_http.sql_over_http_client_conn_threshold,
     };
     let authentication_config = AuthenticationConfig {
-        thread_pool,
         scram_protocol_timeout: args.scram_protocol_timeout,
         rate_limiter_enabled: args.auth_rate_limit_enabled,
         rate_limiter: AuthRateLimiter::new(args.auth_rate_limit.clone()),

proxy/src/config.rs

@@ -2,7 +2,6 @@ use crate::{
     auth::{self, backend::AuthRateLimiter},
     console::locks::ApiLocks,
     rate_limiter::RateBucketInfo,
-    scram::threadpool::ThreadPool,
     serverless::{cancel_set::CancelSet, GlobalConnPoolOptions},
     Host,
 };
@@ -62,7 +61,6 @@ pub struct HttpConfig {
 }
 
 pub struct AuthenticationConfig {
-    pub thread_pool: Arc<ThreadPool>,
     pub scram_protocol_timeout: tokio::time::Duration,
     pub rate_limiter_enabled: bool,
     pub rate_limiter: AuthRateLimiter,

proxy/src/context/parquet.rs

@@ -307,7 +307,7 @@ where
 }
 
 async fn upload_parquet(
-    mut w: SerializedFileWriter<Writer<BytesMut>>,
+    w: SerializedFileWriter<Writer<BytesMut>>,
     len: i64,
     storage: &GenericRemoteStorage,
 ) -> anyhow::Result<Writer<BytesMut>> {
@@ -319,15 +319,11 @@ async fn upload_parquet(
 
     // I don't know how compute intensive this is, although it probably isn't much... better be safe than sorry.
     // finish method only available on the fork: https://github.com/apache/arrow-rs/issues/5253
-    let (mut buffer, metadata) =
-        tokio::task::spawn_blocking(move || -> parquet::errors::Result<_> {
-            let metadata = w.finish()?;
-            let buffer = std::mem::take(w.inner_mut().get_mut());
-            Ok((buffer, metadata))
-        })
+    let (writer, metadata) = tokio::task::spawn_blocking(move || w.finish())
         .await
         .unwrap()?;
 
+    let mut buffer = writer.into_inner();
     let data = buffer.split().freeze();
 
     let compression = len as f64 / len_uncompressed as f64;
@@ -355,7 +351,7 @@ async fn upload_parquet(
         "{year:04}/{month:02}/{day:02}/{hour:02}/requests_{id}.parquet"
     ))?;
     let cancel = CancellationToken::new();
-    let maybe_err = backoff::retry(
+    backoff::retry(
         || async {
             let stream = futures::stream::once(futures::future::ready(Ok(data.clone())));
             storage
@@ -372,12 +368,7 @@ async fn upload_parquet(
     .await
     .ok_or_else(|| anyhow::Error::new(TimeoutOrCancel::Cancel))
     .and_then(|x| x)
-    .context("request_data_upload")
-    .err();
-
-    if let Some(err) = maybe_err {
-        tracing::warn!(%id, %err, "failed to upload request data");
-    }
+    .context("request_data_upload")?;
 
     Ok(buffer.writer())
 }
@@ -483,11 +474,10 @@ mod tests {
         RequestData {
             session_id: uuid::Builder::from_random_bytes(rng.gen()).into_uuid(),
             peer_addr: Ipv4Addr::from(rng.gen::<[u8; 4]>()).to_string(),
-            timestamp: chrono::DateTime::from_timestamp_millis(
+            timestamp: chrono::NaiveDateTime::from_timestamp_millis(
                 rng.gen_range(1703862754..1803862754),
             )
-            .unwrap()
-            .naive_utc(),
+            .unwrap(),
             application_name: Some("test".to_owned()),
             username: Some(hex::encode(rng.gen::<[u8; 4]>())),
             endpoint_id: Some(hex::encode(rng.gen::<[u8; 16]>())),
@@ -570,15 +560,15 @@ mod tests {
         assert_eq!(
             file_stats,
             [
-                (1315314, 3, 6000),
-                (1315307, 3, 6000),
-                (1315367, 3, 6000),
-                (1315324, 3, 6000),
-                (1315454, 3, 6000),
-                (1315296, 3, 6000),
-                (1315088, 3, 6000),
-                (1315324, 3, 6000),
-                (438713, 1, 2000)
+                (1315008, 3, 6000),
+                (1315001, 3, 6000),
+                (1315061, 3, 6000),
+                (1315018, 3, 6000),
+                (1315148, 3, 6000),
+                (1314990, 3, 6000),
+                (1314782, 3, 6000),
+                (1315018, 3, 6000),
+                (438575, 1, 2000)
             ]
         );
 
@@ -608,11 +598,11 @@ mod tests {
         assert_eq!(
             file_stats,
             [
-                (1222212, 5, 10000),
-                (1228362, 5, 10000),
-                (1230156, 5, 10000),
-                (1229518, 5, 10000),
-                (1220796, 5, 10000)
+                (1221738, 5, 10000),
+                (1227888, 5, 10000),
+                (1229682, 5, 10000),
+                (1229044, 5, 10000),
+                (1220322, 5, 10000)
             ]
         );
 
@@ -644,11 +634,11 @@ mod tests {
         assert_eq!(
             file_stats,
             [
-                (1207859, 5, 10000),
-                (1207590, 5, 10000),
-                (1207883, 5, 10000),
-                (1207871, 5, 10000),
-                (1208126, 5, 10000)
+                (1207385, 5, 10000),
+                (1207116, 5, 10000),
+                (1207409, 5, 10000),
+                (1207397, 5, 10000),
+                (1207652, 5, 10000)
             ]
         );
 
@@ -673,15 +663,15 @@ mod tests {
         assert_eq!(
             file_stats,
             [
-                (1315314, 3, 6000),
-                (1315307, 3, 6000),
-                (1315367, 3, 6000),
-                (1315324, 3, 6000),
-                (1315454, 3, 6000),
-                (1315296, 3, 6000),
-                (1315088, 3, 6000),
-                (1315324, 3, 6000),
-                (438713, 1, 2000)
+                (1315008, 3, 6000),
+                (1315001, 3, 6000),
+                (1315061, 3, 6000),
+                (1315018, 3, 6000),
+                (1315148, 3, 6000),
+                (1314990, 3, 6000),
+                (1314782, 3, 6000),
+                (1315018, 3, 6000),
+                (438575, 1, 2000)
             ]
         );
 
@@ -718,7 +708,7 @@ mod tests {
         // files are smaller than the size threshold, but they took too long to fill so were flushed early
         assert_eq!(
             file_stats,
-            [(659462, 2, 3001), (659176, 2, 3000), (658972, 2, 2999)]
+            [(659240, 2, 3001), (658954, 2, 3000), (658750, 2, 2999)]
         );
 
         tmpdir.close().unwrap();

proxy/src/metrics.rs

@@ -1,11 +1,11 @@
-use std::sync::{Arc, OnceLock};
+use std::sync::OnceLock;
 
 use lasso::ThreadedRodeo;
 use measured::{
-    label::{FixedCardinalitySet, LabelName, LabelSet, LabelValue, StaticLabelSet},
+    label::StaticLabelSet,
     metric::{histogram::Thresholds, name::MetricName},
-    Counter, CounterVec, FixedCardinalityLabel, Gauge, GaugeVec, Histogram, HistogramVec,
-    LabelGroup, MetricGroup,
+    Counter, CounterVec, FixedCardinalityLabel, Gauge, Histogram, HistogramVec, LabelGroup,
+    MetricGroup,
 };
 use metrics::{CounterPairAssoc, CounterPairVec, HyperLogLog, HyperLogLogVec};
 
@@ -14,36 +14,26 @@ use tokio::time::{self, Instant};
 use crate::console::messages::ColdStartInfo;
 
 #[derive(MetricGroup)]
-#[metric(new(thread_pool: Arc<ThreadPoolMetrics>))]
 pub struct Metrics {
     #[metric(namespace = "proxy")]
-    #[metric(init = ProxyMetrics::new(thread_pool))]
     pub proxy: ProxyMetrics,
 
     #[metric(namespace = "wake_compute_lock")]
     pub wake_compute_lock: ApiLockMetrics,
 }
 
-static SELF: OnceLock<Metrics> = OnceLock::new();
 impl Metrics {
-    pub fn install(thread_pool: Arc<ThreadPoolMetrics>) {
-        SELF.set(Metrics::new(thread_pool))
-            .ok()
-            .expect("proxy metrics must not be installed more than once");
-    }
-
     pub fn get() -> &'static Self {
-        #[cfg(test)]
-        return SELF.get_or_init(|| Metrics::new(Arc::new(ThreadPoolMetrics::new(0))));
-
-        #[cfg(not(test))]
-        SELF.get()
-            .expect("proxy metrics must be installed by the main() function")
+        static SELF: OnceLock<Metrics> = OnceLock::new();
+        SELF.get_or_init(|| Metrics {
+            proxy: ProxyMetrics::default(),
+            wake_compute_lock: ApiLockMetrics::new(),
+        })
     }
 }
 
 #[derive(MetricGroup)]
-#[metric(new(thread_pool: Arc<ThreadPoolMetrics>))]
+#[metric(new())]
 pub struct ProxyMetrics {
     #[metric(flatten)]
     pub db_connections: CounterPairVec<NumDbConnectionsGauge>,
@@ -139,10 +129,6 @@ pub struct ProxyMetrics {
 
     #[metric(namespace = "connect_compute_lock")]
     pub connect_compute_lock: ApiLockMetrics,
-
-    #[metric(namespace = "scram_pool")]
-    #[metric(init = thread_pool)]
-    pub scram_pool: Arc<ThreadPoolMetrics>,
 }
 
 #[derive(MetricGroup)]
@@ -160,6 +146,12 @@ pub struct ApiLockMetrics {
     pub semaphore_acquire_seconds: Histogram<16>,
 }
 
+impl Default for ProxyMetrics {
+    fn default() -> Self {
+        Self::new()
+    }
+}
+
 impl Default for ApiLockMetrics {
     fn default() -> Self {
         Self::new()
@@ -561,52 +553,3 @@ pub enum RedisEventsCount {
     PasswordUpdate,
     AllowedIpsUpdate,
 }
-
-pub struct ThreadPoolWorkers(usize);
-pub struct ThreadPoolWorkerId(pub usize);
-
-impl LabelValue for ThreadPoolWorkerId {
-    fn visit<V: measured::label::LabelVisitor>(&self, v: V) -> V::Output {
-        v.write_int(self.0 as i64)
-    }
-}
-
-impl LabelGroup for ThreadPoolWorkerId {
-    fn visit_values(&self, v: &mut impl measured::label::LabelGroupVisitor) {
-        v.write_value(LabelName::from_str("worker"), self);
-    }
-}
-
-impl LabelSet for ThreadPoolWorkers {
-    type Value<'a> = ThreadPoolWorkerId;
-
-    fn dynamic_cardinality(&self) -> Option<usize> {
-        Some(self.0)
-    }
-
-    fn encode(&self, value: Self::Value<'_>) -> Option<usize> {
-        (value.0 < self.0).then_some(value.0)
-    }
-
-    fn decode(&self, value: usize) -> Self::Value<'_> {
-        ThreadPoolWorkerId(value)
-    }
-}
-
-impl FixedCardinalitySet for ThreadPoolWorkers {
-    fn cardinality(&self) -> usize {
-        self.0
-    }
-}
-
-#[derive(MetricGroup)]
-#[metric(new(workers: usize))]
-pub struct ThreadPoolMetrics {
-    pub injector_queue_depth: Gauge,
-    #[metric(init = GaugeVec::with_label_set(ThreadPoolWorkers(workers)))]
-    pub worker_queue_depth: GaugeVec<ThreadPoolWorkers>,
-    #[metric(init = CounterVec::with_label_set(ThreadPoolWorkers(workers)))]
-    pub worker_task_turns_total: CounterVec<ThreadPoolWorkers>,
-    #[metric(init = CounterVec::with_label_set(ThreadPoolWorkers(workers)))]
-    pub worker_task_skips_total: CounterVec<ThreadPoolWorkers>,
-}

proxy/src/scram.rs

@@ -6,14 +6,11 @@
 //! * <https://github.com/postgres/postgres/blob/94226d4506e66d6e7cbf4b391f1e7393c1962841/src/backend/libpq/auth-scram.c>
 //! * <https://github.com/postgres/postgres/blob/94226d4506e66d6e7cbf4b391f1e7393c1962841/src/interfaces/libpq/fe-auth-scram.c>
 
-mod countmin;
 mod exchange;
 mod key;
 mod messages;
-mod pbkdf2;
 mod secret;
 mod signature;
-pub mod threadpool;
 
 pub use exchange::{exchange, Exchange};
 pub use key::ScramKey;
@@ -59,13 +56,9 @@ fn sha256<'a>(parts: impl IntoIterator<Item = &'a [u8]>) -> [u8; 32] {
 
 #[cfg(test)]
 mod tests {
-    use crate::{
-        intern::EndpointIdInt,
-        sasl::{Mechanism, Step},
-        EndpointId,
-    };
+    use crate::sasl::{Mechanism, Step};
 
-    use super::{threadpool::ThreadPool, Exchange, ServerSecret};
+    use super::{Exchange, ServerSecret};
 
     #[test]
     fn snapshot() {
@@ -119,13 +112,8 @@ mod tests {
     }
 
     async fn run_round_trip_test(server_password: &str, client_password: &str) {
-        let pool = ThreadPool::new(1);
-
-        let ep = EndpointId::from("foo");
-        let ep = EndpointIdInt::from(ep);
-
         let scram_secret = ServerSecret::build(server_password).await.unwrap();
-        let outcome = super::exchange(&pool, ep, &scram_secret, client_password.as_bytes())
+        let outcome = super::exchange(&scram_secret, client_password.as_bytes())
             .await
             .unwrap();
 

proxy/src/scram/countmin.rs

@@ -1,173 +0,0 @@
-use std::hash::Hash;
-
-/// estimator of hash jobs per second.
-/// <https://en.wikipedia.org/wiki/Count%E2%80%93min_sketch>
-pub struct CountMinSketch {
-    // one for each depth
-    hashers: Vec<ahash::RandomState>,
-    width: usize,
-    depth: usize,
-    // buckets, width*depth
-    buckets: Vec<u32>,
-}
-
-impl CountMinSketch {
-    /// Given parameters (ε, δ),
-    ///   set width = ceil(e/ε)
-    ///   set depth = ceil(ln(1/δ))
-    ///
-    /// guarantees:
-    /// actual <= estimate
-    /// estimate <= actual + ε * N with probability 1 - δ
-    /// where N is the cardinality of the stream
-    pub fn with_params(epsilon: f64, delta: f64) -> Self {
-        CountMinSketch::new(
-            (std::f64::consts::E / epsilon).ceil() as usize,
-            (1.0_f64 / delta).ln().ceil() as usize,
-        )
-    }
-
-    fn new(width: usize, depth: usize) -> Self {
-        Self {
-            #[cfg(test)]
-            hashers: (0..depth)
-                .map(|i| {
-                    // digits of pi for good randomness
-                    ahash::RandomState::with_seeds(
-                        314159265358979323,
-                        84626433832795028,
-                        84197169399375105,
-                        82097494459230781 + i as u64,
-                    )
-                })
-                .collect(),
-            #[cfg(not(test))]
-            hashers: (0..depth).map(|_| ahash::RandomState::new()).collect(),
-            width,
-            depth,
-            buckets: vec![0; width * depth],
-        }
-    }
-
-    pub fn inc_and_return<T: Hash>(&mut self, t: &T, x: u32) -> u32 {
-        let mut min = u32::MAX;
-        for row in 0..self.depth {
-            let col = (self.hashers[row].hash_one(t) as usize) % self.width;
-
-            let row = &mut self.buckets[row * self.width..][..self.width];
-            row[col] = row[col].saturating_add(x);
-            min = std::cmp::min(min, row[col]);
-        }
-        min
-    }
-
-    pub fn reset(&mut self) {
-        self.buckets.clear();
-        self.buckets.resize(self.width * self.depth, 0);
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use rand::{rngs::StdRng, seq::SliceRandom, Rng, SeedableRng};
-
-    use super::CountMinSketch;
-
-    fn eval_precision(n: usize, p: f64, q: f64) -> usize {
-        // fixed value of phi for consistent test
-        let mut rng = StdRng::seed_from_u64(16180339887498948482);
-
-        #[allow(non_snake_case)]
-        let mut N = 0;
-
-        let mut ids = vec![];
-
-        for _ in 0..n {
-            // number of insert operations
-            let n = rng.gen_range(1..100);
-            // number to insert at once
-            let m = rng.gen_range(1..4096);
-
-            let id = uuid::Builder::from_random_bytes(rng.gen()).into_uuid();
-            ids.push((id, n, m));
-
-            // N = sum(actual)
-            N += n * m;
-        }
-
-        // q% of counts will be within p of the actual value
-        let mut sketch = CountMinSketch::with_params(p / N as f64, 1.0 - q);
-
-        dbg!(sketch.buckets.len());
-
-        // insert a bunch of entries in a random order
-        let mut ids2 = ids.clone();
-        while !ids2.is_empty() {
-            ids2.shuffle(&mut rng);
-
-            let mut i = 0;
-            while i < ids2.len() {
-                sketch.inc_and_return(&ids2[i].0, ids2[i].1);
-                ids2[i].2 -= 1;
-                if ids2[i].2 == 0 {
-                    ids2.remove(i);
-                } else {
-                    i += 1;
-                }
-            }
-        }
-
-        let mut within_p = 0;
-        for (id, n, m) in ids {
-            let actual = n * m;
-            let estimate = sketch.inc_and_return(&id, 0);
-
-            // This estimate has the guarantee that actual <= estimate
-            assert!(actual <= estimate);
-
-            // This estimate has the guarantee that estimate <= actual + εN with probability 1 - δ.
-            // ε = p / N, δ = 1 - q;
-            // therefore, estimate <= actual + p with probability q.
-            if estimate as f64 <= actual as f64 + p {
-                within_p += 1;
-            }
-        }
-        within_p
-    }
-
-    #[test]
-    fn precision() {
-        assert_eq!(eval_precision(100, 100.0, 0.99), 100);
-        assert_eq!(eval_precision(1000, 100.0, 0.99), 1000);
-        assert_eq!(eval_precision(100, 4096.0, 0.99), 100);
-        assert_eq!(eval_precision(1000, 4096.0, 0.99), 1000);
-
-        // seems to be more precise than the literature indicates?
-        // probably numbers are too small to truly represent the probabilities.
-        assert_eq!(eval_precision(100, 4096.0, 0.90), 100);
-        assert_eq!(eval_precision(1000, 4096.0, 0.90), 1000);
-        assert_eq!(eval_precision(100, 4096.0, 0.1), 98);
-        assert_eq!(eval_precision(1000, 4096.0, 0.1), 991);
-    }
-
-    // returns memory usage in bytes, and the time complexity per insert.
-    fn eval_cost(p: f64, q: f64) -> (usize, usize) {
-        #[allow(non_snake_case)]
-        // N = sum(actual)
-        // Let's assume 1021 samples, all of 4096
-        let N = 1021 * 4096;
-        let sketch = CountMinSketch::with_params(p / N as f64, 1.0 - q);
-
-        let memory = std::mem::size_of::<u32>() * sketch.buckets.len();
-        let time = sketch.depth;
-        (memory, time)
-    }
-
-    #[test]
-    fn memory_usage() {
-        assert_eq!(eval_cost(100.0, 0.99), (2273580, 5));
-        assert_eq!(eval_cost(4096.0, 0.99), (55520, 5));
-        assert_eq!(eval_cost(4096.0, 0.90), (33312, 3));
-        assert_eq!(eval_cost(4096.0, 0.1), (11104, 1));
-    }
-}

proxy/src/scram/exchange.rs

@@ -4,17 +4,15 @@ use std::convert::Infallible;
 
 use hmac::{Hmac, Mac};
 use sha2::Sha256;
+use tokio::task::yield_now;
 
 use super::messages::{
     ClientFinalMessage, ClientFirstMessage, OwnedServerFirstMessage, SCRAM_RAW_NONCE_LEN,
 };
-use super::pbkdf2::Pbkdf2;
 use super::secret::ServerSecret;
 use super::signature::SignatureBuilder;
-use super::threadpool::ThreadPool;
 use super::ScramKey;
 use crate::config;
-use crate::intern::EndpointIdInt;
 use crate::sasl::{self, ChannelBinding, Error as SaslError};
 
 /// The only channel binding mode we currently support.
@@ -76,18 +74,37 @@ impl<'a> Exchange<'a> {
     }
 }
 
+// copied from <https://github.com/neondatabase/rust-postgres/blob/20031d7a9ee1addeae6e0968e3899ae6bf01cee2/postgres-protocol/src/authentication/sasl.rs#L36-L61>
+async fn pbkdf2(str: &[u8], salt: &[u8], iterations: u32) -> [u8; 32] {
+    let hmac = Hmac::<Sha256>::new_from_slice(str).expect("HMAC is able to accept all key sizes");
+    let mut prev = hmac
+        .clone()
+        .chain_update(salt)
+        .chain_update(1u32.to_be_bytes())
+        .finalize()
+        .into_bytes();
+
+    let mut hi = prev;
+
+    for i in 1..iterations {
+        prev = hmac.clone().chain_update(prev).finalize().into_bytes();
+
+        for (hi, prev) in hi.iter_mut().zip(prev) {
+            *hi ^= prev;
+        }
+        // yield every ~250us
+        // hopefully reduces tail latencies
+        if i % 1024 == 0 {
+            yield_now().await
+        }
+    }
+
+    hi.into()
+}
+
 // copied from <https://github.com/neondatabase/rust-postgres/blob/20031d7a9ee1addeae6e0968e3899ae6bf01cee2/postgres-protocol/src/authentication/sasl.rs#L236-L248>
-async fn derive_client_key(
-    pool: &ThreadPool,
-    endpoint: EndpointIdInt,
-    password: &[u8],
-    salt: &[u8],
-    iterations: u32,
-) -> ScramKey {
-    let salted_password = pool
-        .spawn_job(endpoint, Pbkdf2::start(password, salt, iterations))
-        .await
-        .expect("job should not be cancelled");
+async fn derive_client_key(password: &[u8], salt: &[u8], iterations: u32) -> ScramKey {
+    let salted_password = pbkdf2(password, salt, iterations).await;
 
     let make_key = |name| {
         let key = Hmac::<Sha256>::new_from_slice(&salted_password)
@@ -102,13 +119,11 @@ async fn derive_client_key(
 }
 
 pub async fn exchange(
-    pool: &ThreadPool,
-    endpoint: EndpointIdInt,
     secret: &ServerSecret,
     password: &[u8],
 ) -> sasl::Result<sasl::Outcome<super::ScramKey>> {
     let salt = base64::decode(&secret.salt_base64)?;
-    let client_key = derive_client_key(pool, endpoint, password, &salt, secret.iterations).await;
+    let client_key = derive_client_key(password, &salt, secret.iterations).await;
 
     if secret.is_password_invalid(&client_key).into() {
         Ok(sasl::Outcome::Failure("password doesn't match"))

proxy/src/scram/pbkdf2.rs

@@ -1,89 +0,0 @@
-use hmac::{
-    digest::{consts::U32, generic_array::GenericArray},
-    Hmac, Mac,
-};
-use sha2::Sha256;
-
-pub struct Pbkdf2 {
-    hmac: Hmac<Sha256>,
-    prev: GenericArray<u8, U32>,
-    hi: GenericArray<u8, U32>,
-    iterations: u32,
-}
-
-// inspired from <https://github.com/neondatabase/rust-postgres/blob/20031d7a9ee1addeae6e0968e3899ae6bf01cee2/postgres-protocol/src/authentication/sasl.rs#L36-L61>
-impl Pbkdf2 {
-    pub fn start(str: &[u8], salt: &[u8], iterations: u32) -> Self {
-        let hmac =
-            Hmac::<Sha256>::new_from_slice(str).expect("HMAC is able to accept all key sizes");
-
-        let prev = hmac
-            .clone()
-            .chain_update(salt)
-            .chain_update(1u32.to_be_bytes())
-            .finalize()
-            .into_bytes();
-
-        Self {
-            hmac,
-            // one consumed for the hash above
-            iterations: iterations - 1,
-            hi: prev,
-            prev,
-        }
-    }
-
-    pub fn cost(&self) -> u32 {
-        (self.iterations).clamp(0, 4096)
-    }
-
-    pub fn turn(&mut self) -> std::task::Poll<[u8; 32]> {
-        let Self {
-            hmac,
-            prev,
-            hi,
-            iterations,
-        } = self;
-
-        // only do 4096 iterations per turn before sharing the thread for fairness
-        let n = (*iterations).clamp(0, 4096);
-        for _ in 0..n {
-            *prev = hmac.clone().chain_update(*prev).finalize().into_bytes();
-
-            for (hi, prev) in hi.iter_mut().zip(*prev) {
-                *hi ^= prev;
-            }
-        }
-
-        *iterations -= n;
-        if *iterations == 0 {
-            std::task::Poll::Ready((*hi).into())
-        } else {
-            std::task::Poll::Pending
-        }
-    }
-}
-
-#[cfg(test)]
-mod tests {
-    use super::Pbkdf2;
-    use pbkdf2::pbkdf2_hmac_array;
-    use sha2::Sha256;
-
-    #[test]
-    fn works() {
-        let salt = b"sodium chloride";
-        let pass = b"Ne0n_!5_50_C007";
-
-        let mut job = Pbkdf2::start(pass, salt, 600000);
-        let hash = loop {
-            let std::task::Poll::Ready(hash) = job.turn() else {
-                continue;
-            };
-            break hash;
-        };
-
-        let expected = pbkdf2_hmac_array::<Sha256, 32>(pass, salt, 600000);
-        assert_eq!(hash, expected)
-    }
-}

proxy/src/scram/threadpool.rs

@@ -1,321 +0,0 @@
-//! Custom threadpool implementation for password hashing.
-//!
-//! Requirements:
-//! 1. Fairness per endpoint.
-//! 2. Yield support for high iteration counts.
-
-use std::sync::{
-    atomic::{AtomicU64, Ordering},
-    Arc,
-};
-
-use crossbeam_deque::{Injector, Stealer, Worker};
-use itertools::Itertools;
-use parking_lot::{Condvar, Mutex};
-use rand::Rng;
-use rand::{rngs::SmallRng, SeedableRng};
-use tokio::sync::oneshot;
-
-use crate::{
-    intern::EndpointIdInt,
-    metrics::{ThreadPoolMetrics, ThreadPoolWorkerId},
-    scram::countmin::CountMinSketch,
-};
-
-use super::pbkdf2::Pbkdf2;
-
-pub struct ThreadPool {
-    queue: Injector<JobSpec>,
-    stealers: Vec<Stealer<JobSpec>>,
-    parkers: Vec<(Condvar, Mutex<ThreadState>)>,
-    /// bitpacked representation.
-    /// lower 8 bits = number of sleeping threads
-    /// next 8 bits = number of idle threads (searching for work)
-    counters: AtomicU64,
-
-    pub metrics: Arc<ThreadPoolMetrics>,
-}
-
-#[derive(PartialEq)]
-enum ThreadState {
-    Parked,
-    Active,
-}
-
-impl ThreadPool {
-    pub fn new(n_workers: u8) -> Arc<Self> {
-        let workers = (0..n_workers).map(|_| Worker::new_fifo()).collect_vec();
-        let stealers = workers.iter().map(|w| w.stealer()).collect_vec();
-
-        let parkers = (0..n_workers)
-            .map(|_| (Condvar::new(), Mutex::new(ThreadState::Active)))
-            .collect_vec();
-
-        let pool = Arc::new(Self {
-            queue: Injector::new(),
-            stealers,
-            parkers,
-            // threads start searching for work
-            counters: AtomicU64::new((n_workers as u64) << 8),
-            metrics: Arc::new(ThreadPoolMetrics::new(n_workers as usize)),
-        });
-
-        for (i, worker) in workers.into_iter().enumerate() {
-            let pool = Arc::clone(&pool);
-            std::thread::spawn(move || thread_rt(pool, worker, i));
-        }
-
-        pool
-    }
-
-    pub fn spawn_job(
-        &self,
-        endpoint: EndpointIdInt,
-        pbkdf2: Pbkdf2,
-    ) -> oneshot::Receiver<[u8; 32]> {
-        let (tx, rx) = oneshot::channel();
-
-        let queue_was_empty = self.queue.is_empty();
-
-        self.metrics.injector_queue_depth.inc();
-        self.queue.push(JobSpec {
-            response: tx,
-            pbkdf2,
-            endpoint,
-        });
-
-        // inspired from <https://github.com/rayon-rs/rayon/blob/3e3962cb8f7b50773bcc360b48a7a674a53a2c77/rayon-core/src/sleep/mod.rs#L242>
-        let counts = self.counters.load(Ordering::SeqCst);
-        let num_awake_but_idle = (counts >> 8) & 0xff;
-        let num_sleepers = counts & 0xff;
-
-        // If the queue is non-empty, then we always wake up a worker
-        // -- clearly the existing idle jobs aren't enough. Otherwise,
-        // check to see if we have enough idle workers.
-        if !queue_was_empty || num_awake_but_idle == 0 {
-            let num_to_wake = Ord::min(1, num_sleepers);
-            self.wake_any_threads(num_to_wake);
-        }
-
-        rx
-    }
-
-    #[cold]
-    fn wake_any_threads(&self, mut num_to_wake: u64) {
-        if num_to_wake > 0 {
-            for i in 0..self.parkers.len() {
-                if self.wake_specific_thread(i) {
-                    num_to_wake -= 1;
-                    if num_to_wake == 0 {
-                        return;
-                    }
-                }
-            }
-        }
-    }
-
-    fn wake_specific_thread(&self, index: usize) -> bool {
-        let (condvar, lock) = &self.parkers[index];
-
-        let mut state = lock.lock();
-        if *state == ThreadState::Parked {
-            condvar.notify_one();
-
-            // When the thread went to sleep, it will have incremented
-            // this value. When we wake it, its our job to decrement
-            // it. We could have the thread do it, but that would
-            // introduce a delay between when the thread was
-            // *notified* and when this counter was decremented. That
-            // might mislead people with new work into thinking that
-            // there are sleeping threads that they should try to
-            // wake, when in fact there is nothing left for them to
-            // do.
-            self.counters.fetch_sub(1, Ordering::SeqCst);
-            *state = ThreadState::Active;
-
-            true
-        } else {
-            false
-        }
-    }
-
-    fn steal(&self, rng: &mut impl Rng, skip: usize, worker: &Worker<JobSpec>) -> Option<JobSpec> {
-        // announce thread as idle
-        self.counters.fetch_add(256, Ordering::SeqCst);
-
-        // try steal from the global queue
-        loop {
-            match self.queue.steal_batch_and_pop(worker) {
-                crossbeam_deque::Steal::Success(job) => {
-                    self.metrics
-                        .injector_queue_depth
-                        .set(self.queue.len() as i64);
-                    // no longer idle
-                    self.counters.fetch_sub(256, Ordering::SeqCst);
-                    return Some(job);
-                }
-                crossbeam_deque::Steal::Retry => continue,
-                crossbeam_deque::Steal::Empty => break,
-            }
-        }
-
-        // try steal from our neighbours
-        loop {
-            let mut retry = false;
-            let start = rng.gen_range(0..self.stealers.len());
-            let job = (start..self.stealers.len())
-                .chain(0..start)
-                .filter(|i| *i != skip)
-                .find_map(
-                    |victim| match self.stealers[victim].steal_batch_and_pop(worker) {
-                        crossbeam_deque::Steal::Success(job) => Some(job),
-                        crossbeam_deque::Steal::Empty => None,
-                        crossbeam_deque::Steal::Retry => {
-                            retry = true;
-                            None
-                        }
-                    },
-                );
-            if job.is_some() {
-                // no longer idle
-                self.counters.fetch_sub(256, Ordering::SeqCst);
-                return job;
-            }
-            if !retry {
-                return None;
-            }
-        }
-    }
-}
-
-fn thread_rt(pool: Arc<ThreadPool>, worker: Worker<JobSpec>, index: usize) {
-    /// interval when we should steal from the global queue
-    /// so that tail latencies are managed appropriately
-    const STEAL_INTERVAL: usize = 61;
-
-    /// How often to reset the sketch values
-    const SKETCH_RESET_INTERVAL: usize = 1021;
-
-    let mut rng = SmallRng::from_entropy();
-
-    // used to determine whether we should temporarily skip tasks for fairness.
-    // 99% of estimates will overcount by no more than 4096 samples
-    let mut sketch = CountMinSketch::with_params(1.0 / (SKETCH_RESET_INTERVAL as f64), 0.01);
-
-    let (condvar, lock) = &pool.parkers[index];
-
-    'wait: loop {
-        // wait for notification of work
-        {
-            let mut lock = lock.lock();
-
-            // queue is empty
-            pool.metrics
-                .worker_queue_depth
-                .set(ThreadPoolWorkerId(index), 0);
-
-            // subtract 1 from idle count, add 1 to sleeping count.
-            pool.counters.fetch_sub(255, Ordering::SeqCst);
-
-            *lock = ThreadState::Parked;
-            condvar.wait(&mut lock);
-        }
-
-        for i in 0.. {
-            let mut job = match worker
-                .pop()
-                .or_else(|| pool.steal(&mut rng, index, &worker))
-            {
-                Some(job) => job,
-                None => continue 'wait,
-            };
-
-            pool.metrics
-                .worker_queue_depth
-                .set(ThreadPoolWorkerId(index), worker.len() as i64);
-
-            // receiver is closed, cancel the task
-            if !job.response.is_closed() {
-                let rate = sketch.inc_and_return(&job.endpoint, job.pbkdf2.cost());
-
-                const P: f64 = 2000.0;
-                // probability decreases as rate increases.
-                // lower probability, higher chance of being skipped
-                //
-                // estimates (rate in terms of 4096 rounds):
-                // rate = 0    => probability = 100%
-                // rate = 10   => probability = 71.3%
-                // rate = 50   => probability = 62.1%
-                // rate = 500  => probability = 52.3%
-                // rate = 1021 => probability = 49.8%
-                //
-                // My expectation is that the pool queue will only begin backing up at ~1000rps
-                // in which case the SKETCH_RESET_INTERVAL represents 1 second. Thus, the rates above
-                // are in requests per second.
-                let probability = P.ln() / (P + rate as f64).ln();
-                if pool.queue.len() > 32 || rng.gen_bool(probability) {
-                    pool.metrics
-                        .worker_task_turns_total
-                        .inc(ThreadPoolWorkerId(index));
-
-                    match job.pbkdf2.turn() {
-                        std::task::Poll::Ready(result) => {
-                            let _ = job.response.send(result);
-                        }
-                        std::task::Poll::Pending => worker.push(job),
-                    }
-                } else {
-                    pool.metrics
-                        .worker_task_skips_total
-                        .inc(ThreadPoolWorkerId(index));
-
-                    // skip for now
-                    worker.push(job)
-                }
-            }
-
-            // if we get stuck with a few long lived jobs in the queue
-            // it's better to try and steal from the queue too for fairness
-            if i % STEAL_INTERVAL == 0 {
-                let _ = pool.queue.steal_batch(&worker);
-            }
-
-            if i % SKETCH_RESET_INTERVAL == 0 {
-                sketch.reset();
-            }
-        }
-    }
-}
-
-struct JobSpec {
-    response: oneshot::Sender<[u8; 32]>,
-    pbkdf2: Pbkdf2,
-    endpoint: EndpointIdInt,
-}
-
-#[cfg(test)]
-mod tests {
-    use crate::EndpointId;
-
-    use super::*;
-
-    #[tokio::test]
-    async fn hash_is_correct() {
-        let pool = ThreadPool::new(1);
-
-        let ep = EndpointId::from("foo");
-        let ep = EndpointIdInt::from(ep);
-
-        let salt = [0x55; 32];
-        let actual = pool
-            .spawn_job(ep, Pbkdf2::start(b"password", &salt, 4096))
-            .await
-            .unwrap();
-
-        let expected = [
-            10, 114, 73, 188, 140, 222, 196, 156, 214, 184, 79, 157, 119, 242, 16, 31, 53, 242,
-            178, 43, 95, 8, 225, 182, 122, 40, 219, 21, 89, 147, 64, 140,
-        ];
-        assert_eq!(actual, expected)
-    }
-}

proxy/src/serverless.rs

@@ -102,7 +102,7 @@ pub async fn task_main(
     let connections = tokio_util::task::task_tracker::TaskTracker::new();
     connections.close(); // allows `connections.wait to complete`
 
-    let server = Builder::new(TokioExecutor::new());
+    let server = Builder::new(hyper_util::rt::TokioExecutor::new());
 
     while let Some(res) = run_until_cancelled(ws_listener.accept(), &cancellation_token).await {
         let (conn, peer_addr) = res.context("could not accept TCP stream")?;
@@ -255,6 +255,7 @@ async fn connection_handler(
                 .in_current_span()
                 .map_ok_or_else(api_error_into_response, |r| r),
             );
+
             async move {
                 let res = handler.await;
                 cancel_request.disarm();
@@ -300,7 +301,7 @@ async fn request_handler(
         .map(|s| s.to_string());
 
     // Check if the request is a websocket upgrade request.
-    if framed_websockets::upgrade::is_upgrade_request(&request) {
+    if hyper_tungstenite::is_upgrade_request(&request) {
         let ctx = RequestMonitoring::new(
             session_id,
             peer_addr,
@@ -311,7 +312,7 @@ async fn request_handler(
         let span = ctx.span.clone();
         info!(parent: &span, "performing websocket upgrade");
 
-        let (response, websocket) = framed_websockets::upgrade::upgrade(&mut request)
+        let (response, websocket) = hyper_tungstenite::upgrade(&mut request, None)
             .map_err(|e| ApiError::BadRequest(e.into()))?;
 
         ws_connections.spawn(
@@ -333,7 +334,7 @@ async fn request_handler(
         );
 
         // Return the response so the spawned future can continue.
-        Ok(response.map(|_: http_body_util::Empty<Bytes>| Full::new(Bytes::new())))
+        Ok(response)
     } else if request.uri().path() == "/sql" && *request.method() == Method::POST {
         let ctx = RequestMonitoring::new(
             session_id,

proxy/src/serverless/backend.rs

@@ -15,7 +15,6 @@ use crate::{
     },
     context::RequestMonitoring,
     error::{ErrorKind, ReportableError, UserFacingError},
-    intern::EndpointIdInt,
     proxy::{connect_compute::ConnectMechanism, retry::ShouldRetry},
     rate_limiter::EndpointRateLimiter,
     Host,
@@ -67,14 +66,8 @@ impl PoolingBackend {
                 return Err(AuthError::auth_failed(&*user_info.user));
             }
         };
-        let ep = EndpointIdInt::from(&conn_info.user_info.endpoint);
-        let auth_outcome = crate::auth::validate_password_and_exchange(
-            &config.thread_pool,
-            ep,
-            &conn_info.password,
-            secret,
-        )
-        .await?;
+        let auth_outcome =
+            crate::auth::validate_password_and_exchange(&conn_info.password, secret).await?;
         let res = match auth_outcome {
             crate::sasl::Outcome::Success(key) => {
                 info!("user successfully authenticated");

proxy/src/serverless/websocket.rs

@@ -7,11 +7,10 @@ use crate::{
     proxy::{handle_client, ClientMode},
     rate_limiter::EndpointRateLimiter,
 };
-use bytes::{Buf, BufMut, Bytes, BytesMut};
-use framed_websockets::{Frame, OpCode, WebSocketServer};
+use bytes::{Buf, Bytes};
 use futures::{Sink, Stream};
-use hyper1::upgrade::OnUpgrade;
-use hyper_util::rt::TokioIo;
+use hyper::upgrade::Upgraded;
+use hyper_tungstenite::{tungstenite::Message, HyperWebsocket, WebSocketStream};
 use pin_project_lite::pin_project;
 
 use std::{
@@ -22,23 +21,25 @@ use std::{
 use tokio::io::{self, AsyncBufRead, AsyncRead, AsyncWrite, ReadBuf};
 use tracing::warn;
 
+// TODO: use `std::sync::Exclusive` once it's stabilized.
+// Tracking issue: https://github.com/rust-lang/rust/issues/98407.
+use sync_wrapper::SyncWrapper;
+
 pin_project! {
     /// This is a wrapper around a [`WebSocketStream`] that
     /// implements [`AsyncRead`] and [`AsyncWrite`].
-    pub struct WebSocketRw<S> {
+    pub struct WebSocketRw<S = Upgraded> {
         #[pin]
-        stream: WebSocketServer<S>,
-        recv: Bytes,
-        send: BytesMut,
+        stream: SyncWrapper<WebSocketStream<S>>,
+        bytes: Bytes,
     }
 }
 
 impl<S> WebSocketRw<S> {
-    pub fn new(stream: WebSocketServer<S>) -> Self {
+    pub fn new(stream: WebSocketStream<S>) -> Self {
         Self {
-            stream,
-            recv: Bytes::new(),
-            send: BytesMut::new(),
+            stream: stream.into(),
+            bytes: Bytes::new(),
         }
     }
 }
@@ -49,24 +50,22 @@ impl<S: AsyncRead + AsyncWrite + Unpin> AsyncWrite for WebSocketRw<S> {
         cx: &mut Context<'_>,
         buf: &[u8],
     ) -> Poll<io::Result<usize>> {
-        let this = self.project();
-        let mut stream = this.stream;
-        this.send.put(buf);
+        let mut stream = self.project().stream.get_pin_mut();
 
         ready!(stream.as_mut().poll_ready(cx).map_err(io_error))?;
-        match stream.as_mut().start_send(Frame::binary(this.send.split())) {
+        match stream.as_mut().start_send(Message::Binary(buf.into())) {
             Ok(()) => Poll::Ready(Ok(buf.len())),
             Err(e) => Poll::Ready(Err(io_error(e))),
         }
     }
 
     fn poll_flush(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
-        let stream = self.project().stream;
+        let stream = self.project().stream.get_pin_mut();
         stream.poll_flush(cx).map_err(io_error)
     }
 
     fn poll_shutdown(self: Pin<&mut Self>, cx: &mut Context<'_>) -> Poll<io::Result<()>> {
-        let stream = self.project().stream;
+        let stream = self.project().stream.get_pin_mut();
         stream.poll_close(cx).map_err(io_error)
     }
 }
@@ -77,10 +76,13 @@ impl<S: AsyncRead + AsyncWrite + Unpin> AsyncRead for WebSocketRw<S> {
         cx: &mut Context<'_>,
         buf: &mut ReadBuf<'_>,
     ) -> Poll<io::Result<()>> {
-        let bytes = ready!(self.as_mut().poll_fill_buf(cx))?;
-        let len = std::cmp::min(bytes.len(), buf.remaining());
-        buf.put_slice(&bytes[..len]);
-        self.consume(len);
+        if buf.remaining() > 0 {
+            let bytes = ready!(self.as_mut().poll_fill_buf(cx))?;
+            let len = std::cmp::min(bytes.len(), buf.remaining());
+            buf.put_slice(&bytes[..len]);
+            self.consume(len);
+        }
+
         Poll::Ready(Ok(()))
     }
 }
@@ -92,27 +94,31 @@ impl<S: AsyncRead + AsyncWrite + Unpin> AsyncBufRead for WebSocketRw<S> {
 
         let mut this = self.project();
         loop {
-            if !this.recv.chunk().is_empty() {
-                let chunk = (*this.recv).chunk();
+            if !this.bytes.chunk().is_empty() {
+                let chunk = (*this.bytes).chunk();
                 return Poll::Ready(Ok(chunk));
             }
 
-            let res = ready!(this.stream.as_mut().poll_next(cx));
+            let res = ready!(this.stream.as_mut().get_pin_mut().poll_next(cx));
             match res.transpose().map_err(io_error)? {
-                Some(message) => match message.opcode {
-                    OpCode::Ping => {}
-                    OpCode::Pong => {}
-                    OpCode::Text => {
+                Some(message) => match message {
+                    Message::Ping(_) => {}
+                    Message::Pong(_) => {}
+                    Message::Text(text) => {
                         // We expect to see only binary messages.
                         let error = "unexpected text message in the websocket";
-                        warn!(length = message.payload.len(), error);
+                        warn!(length = text.len(), error);
                         return Poll::Ready(Err(io_error(error)));
                     }
-                    OpCode::Binary | OpCode::Continuation => {
-                        debug_assert!(this.recv.is_empty());
-                        *this.recv = message.payload.freeze();
+                    Message::Frame(_) => {
+                        // This case is impossible according to Frame's doc.
+                        panic!("unexpected raw frame in the websocket");
                     }
-                    OpCode::Close => return EOF,
+                    Message::Binary(chunk) => {
+                        assert!(this.bytes.is_empty());
+                        *this.bytes = Bytes::from(chunk);
+                    }
+                    Message::Close(_) => return EOF,
                 },
                 None => return EOF,
             }
@@ -120,21 +126,19 @@ impl<S: AsyncRead + AsyncWrite + Unpin> AsyncBufRead for WebSocketRw<S> {
     }
 
     fn consume(self: Pin<&mut Self>, amount: usize) {
-        self.project().recv.advance(amount);
+        self.project().bytes.advance(amount);
     }
 }
 
 pub async fn serve_websocket(
     config: &'static ProxyConfig,
     mut ctx: RequestMonitoring,
-    websocket: OnUpgrade,
+    websocket: HyperWebsocket,
     cancellation_handler: Arc<CancellationHandlerMain>,
     endpoint_rate_limiter: Arc<EndpointRateLimiter>,
     hostname: Option<String>,
 ) -> anyhow::Result<()> {
     let websocket = websocket.await?;
-    let websocket = WebSocketServer::after_handshake(TokioIo::new(websocket));
-
     let conn_gauge = Metrics::get()
         .proxy
         .client_connections
@@ -173,16 +177,15 @@ pub async fn serve_websocket(
 mod tests {
     use std::pin::pin;
 
-    use framed_websockets::WebSocketServer;
     use futures::{SinkExt, StreamExt};
+    use hyper_tungstenite::{
+        tungstenite::{protocol::Role, Message},
+        WebSocketStream,
+    };
     use tokio::{
         io::{duplex, AsyncReadExt, AsyncWriteExt},
         task::JoinSet,
     };
-    use tokio_tungstenite::{
-        tungstenite::{protocol::Role, Message},
-        WebSocketStream,
-    };
 
     use super::WebSocketRw;
 
@@ -207,7 +210,9 @@ mod tests {
         });
 
         js.spawn(async move {
-            let mut rw = pin!(WebSocketRw::new(WebSocketServer::after_handshake(stream2)));
+            let mut rw = pin!(WebSocketRw::new(
+                WebSocketStream::from_raw_socket(stream2, Role::Server, None).await
+            ));
 
             let mut buf = vec![0; 1024];
             let n = rw.read(&mut buf).await.unwrap();

pyproject.toml

@@ -10,7 +10,7 @@ pytest = "^7.4.4"
 psycopg2-binary = "^2.9.6"
 typing-extensions = "^4.6.1"
 PyJWT = {version = "^2.1.0", extras = ["crypto"]}
-requests = "^2.32.0"
+requests = "^2.31.0"
 pytest-xdist = "^3.3.1"
 asyncpg = "^0.29.0"
 aiopg = "^1.4.0"

s3_scrubber/README.md

@@ -9,13 +9,11 @@ and `safekeeper`, and does housekeeping such as cleaning up objects for tenants
 
 #### S3
 
-Do `aws sso login --profile dev` to get the SSO access to the bucket to clean.
-Also, set the following environment variables:
+Do `aws sso login --profile dev` to get the SSO access to the bucket to clean, get the SSO_ACCOUNT_ID for your profile (`cat ~/.aws/config` may help).
 
-- `AWS_PROFILE`: Profile name to use for accessing S3 buckets (e.g. `dev`)
+- `SSO_ACCOUNT_ID`: Credentials id to use for accessing S3 buckets
 - `REGION`: A region where the bucket is located at.
 - `BUCKET`: Bucket name
-- `BUCKET_PREFIX` (optional): Prefix inside the bucket
 
 #### Console API
 
@@ -45,7 +43,7 @@ processing by the `purge-garbage` subcommand.
 
 Example:
 
-`env AWS_PROFILE=dev REGION=eu-west-1 BUCKET=my-dev-bucket CLOUD_ADMIN_API_TOKEN=${NEON_CLOUD_ADMIN_API_STAGING_KEY} CLOUD_ADMIN_API_URL=[url] cargo run --release -- find-garbage --node-kind=pageserver --depth=tenant --output-path=eu-west-1-garbage.json`
+`env SSO_ACCOUNT_ID=123456 REGION=eu-west-1 BUCKET=my-dev-bucket CLOUD_ADMIN_API_TOKEN=${NEON_CLOUD_ADMIN_API_STAGING_KEY} CLOUD_ADMIN_API_URL=[url] cargo run --release -- find-garbage --node-kind=pageserver --depth=tenant --output-path=eu-west-1-garbage.json`
 
 #### `purge-garbage`
 
@@ -61,7 +59,7 @@ to pass them on the command line
 
 Example:
 
-`env AWS_PROFILE=dev cargo run --release -- purge-garbage --node-kind=pageserver --depth=tenant --input-path=eu-west-1-garbage.json`
+`env SSO_ACCOUNT_ID=123456 cargo run --release -- purge-garbage --node-kind=pageserver --depth=tenant --input-path=eu-west-1-garbage.json`
 
 Add the `--delete` argument before `purge-garbage` to enable deletion.  This is intentionally
 not provided inline in the example above to avoid accidents.  Without the `--delete` flag
@@ -74,7 +72,7 @@ Errors are logged to stderr and summary to stdout.
 
 For pageserver:
 ```
-env AWS_PROFILE=dev REGION=eu-west-1 BUCKET=my-dev-bucket CLOUD_ADMIN_API_TOKEN=${NEON_CLOUD_ADMIN_API_STAGING_KEY} CLOUD_ADMIN_API_URL=[url] cargo run --release -- scan-metadata --node-kind pageserver
+env SSO_ACCOUNT_ID=123456 REGION=eu-west-1 BUCKET=my-dev-bucket CLOUD_ADMIN_API_TOKEN=${NEON_CLOUD_ADMIN_API_STAGING_KEY} CLOUD_ADMIN_API_URL=[url] cargo run --release -- scan-metadata --node-kind pageserver
 
 Timelines: 31106
 With errors: 3

s3_scrubber/src/lib.rs

@@ -200,15 +200,30 @@ impl RootTarget {
 }
 
 #[derive(Debug, Clone, Serialize, Deserialize)]
-#[serde(deny_unknown_fields)]
 pub struct BucketConfig {
     pub region: String,
     pub bucket: String,
     pub prefix_in_bucket: Option<String>,
+
+    /// Use SSO if this is set, else rely on AWS_* environment vars
+    pub sso_account_id: Option<String>,
+}
+
+impl Display for BucketConfig {
+    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+        write!(
+            f,
+            "{}/{}/{}",
+            self.sso_account_id.as_deref().unwrap_or("<none>"),
+            self.region,
+            self.bucket
+        )
+    }
 }
 
 impl BucketConfig {
     pub fn from_env() -> anyhow::Result<Self> {
+        let sso_account_id = env::var("SSO_ACCOUNT_ID").ok();
         let region = env::var("REGION").context("'REGION' param retrieval")?;
         let bucket = env::var("BUCKET").context("'BUCKET' param retrieval")?;
         let prefix_in_bucket = env::var("BUCKET_PREFIX").ok();
@@ -217,6 +232,7 @@ impl BucketConfig {
             region,
             bucket,
             prefix_in_bucket,
+            sso_account_id,
         })
     }
 }
@@ -260,7 +276,7 @@ pub fn init_logging(file_name: &str) -> WorkerGuard {
     guard
 }
 
-pub fn init_s3_client(bucket_region: Region) -> Client {
+pub fn init_s3_client(account_id: Option<String>, bucket_region: Region) -> Client {
     let credentials_provider = {
         // uses "AWS_ACCESS_KEY_ID", "AWS_SECRET_ACCESS_KEY"
         let chain = CredentialsProviderChain::first_try(
@@ -274,7 +290,7 @@ pub fn init_s3_client(bucket_region: Region) -> Client {
         );
 
         // Use SSO if we were given an account ID
-        match std::env::var("SSO_ACCOUNT_ID").ok() {
+        match account_id {
             Some(sso_account) => chain.or_else(
                 "sso",
                 SsoCredentialsProvider::builder()
@@ -318,7 +334,7 @@ fn init_remote(
 ) -> anyhow::Result<(Arc<Client>, RootTarget)> {
     let bucket_region = Region::new(bucket_config.region);
     let delimiter = "/".to_string();
-    let s3_client = Arc::new(init_s3_client(bucket_region));
+    let s3_client = Arc::new(init_s3_client(bucket_config.sso_account_id, bucket_region));
 
     let s3_root = match node_kind {
         NodeKind::Pageserver => RootTarget::Pageserver(S3Target {

safekeeper/src/bin/safekeeper.rs

@@ -20,6 +20,7 @@ use std::str::FromStr;
 use std::sync::Arc;
 use std::time::Duration;
 use storage_broker::Uri;
+use tokio::sync::mpsc;
 
 use tracing::*;
 use utils::pid_file;
@@ -29,13 +30,13 @@ use safekeeper::defaults::{
     DEFAULT_HEARTBEAT_TIMEOUT, DEFAULT_HTTP_LISTEN_ADDR, DEFAULT_MAX_OFFLOADER_LAG_BYTES,
     DEFAULT_PARTIAL_BACKUP_TIMEOUT, DEFAULT_PG_LISTEN_ADDR,
 };
-use safekeeper::remove_wal;
 use safekeeper::wal_service;
 use safekeeper::GlobalTimelines;
 use safekeeper::SafeKeeperConf;
 use safekeeper::{broker, WAL_SERVICE_RUNTIME};
 use safekeeper::{control_file, BROKER_RUNTIME};
 use safekeeper::{http, WAL_REMOVER_RUNTIME};
+use safekeeper::{remove_wal, WAL_BACKUP_RUNTIME};
 use safekeeper::{wal_backup, HTTP_RUNTIME};
 use storage_broker::DEFAULT_ENDPOINT;
 use utils::auth::{JwtAuth, Scope, SwappableJwtAuth};
@@ -376,6 +377,8 @@ async fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
     let timeline_collector = safekeeper::metrics::TimelineCollector::new();
     metrics::register_internal(Box::new(timeline_collector))?;
 
+    let (wal_backup_launcher_tx, wal_backup_launcher_rx) = mpsc::channel(100);
+
     wal_backup::init_remote_storage(&conf);
 
     // Keep handles to main tasks to die if any of them disappears.
@@ -388,9 +391,19 @@ async fn start_safekeeper(conf: SafeKeeperConf) -> Result<()> {
     let current_thread_rt = conf
         .current_thread_runtime
         .then(|| Handle::try_current().expect("no runtime in main"));
+    let conf_ = conf.clone();
+    let wal_backup_handle = current_thread_rt
+        .as_ref()
+        .unwrap_or_else(|| WAL_BACKUP_RUNTIME.handle())
+        .spawn(wal_backup::wal_backup_launcher_task_main(
+            conf_,
+            wal_backup_launcher_rx,
+        ))
+        .map(|res| ("WAL backup launcher".to_owned(), res));
+    tasks_handles.push(Box::pin(wal_backup_handle));
 
     // Load all timelines from disk to memory.
-    GlobalTimelines::init(conf.clone()).await?;
+    GlobalTimelines::init(conf.clone(), wal_backup_launcher_tx).await?;
 
     let conf_ = conf.clone();
     // Run everything in current thread rt, if asked.

safekeeper/src/broker.rs

@@ -46,8 +46,6 @@ async fn push_loop(conf: SafeKeeperConf) -> anyhow::Result<()> {
         return Ok(());
     }
 
-    let active_timelines_set = GlobalTimelines::get_global_broker_active_set();
-
     let mut client =
         storage_broker::connect(conf.broker_endpoint.clone(), conf.broker_keepalive_interval)?;
     let push_interval = Duration::from_millis(PUSH_INTERVAL_MSEC);
@@ -59,9 +57,15 @@ async fn push_loop(conf: SafeKeeperConf) -> anyhow::Result<()> {
             // sensitive and there is no risk of deadlock as we don't await while
             // lock is held.
             let now = Instant::now();
-            let all_tlis = active_timelines_set.get_all();
+            let all_tlis = GlobalTimelines::get_all();
             let mut n_pushed_tlis = 0;
             for tli in &all_tlis {
+                // filtering alternative futures::stream::iter(all_tlis)
+                //   .filter(|tli| {let tli = tli.clone(); async move { tli.is_active().await}}).collect::<Vec<_>>().await;
+                // doesn't look better, and I'm not sure how to do that without collect.
+                if !tli.is_active().await {
+                    continue;
+                }
                 let sk_info = tli.get_safekeeper_info(&conf).await;
                 yield sk_info;
                 BROKER_PUSHED_UPDATES.inc();
@@ -86,7 +90,6 @@ async fn push_loop(conf: SafeKeeperConf) -> anyhow::Result<()> {
 }
 
 /// Subscribe and fetch all the interesting data from the broker.
-#[instrument(name = "broker pull", skip_all)]
 async fn pull_loop(conf: SafeKeeperConf, stats: Arc<BrokerStats>) -> Result<()> {
     let mut client = storage_broker::connect(conf.broker_endpoint, conf.broker_keepalive_interval)?;
 
@@ -183,7 +186,6 @@ async fn discover_loop(conf: SafeKeeperConf, stats: Arc<BrokerStats>) -> Result<
                         commit_lsn: sk_info.commit_lsn,
                         safekeeper_connstr: sk_info.safekeeper_connstr,
                         availability_zone: sk_info.availability_zone,
-                        standby_horizon: 0,
                     };
 
                     // note this is a blocking call
@@ -317,7 +319,7 @@ async fn task_stats(stats: Arc<BrokerStats>) {
 
                 let now = BrokerStats::now_millis();
                 if now > last_pulled && now - last_pulled > warn_duration.as_millis() as u64 {
-                    let ts = chrono::DateTime::from_timestamp_millis(last_pulled as i64).expect("invalid timestamp");
+                    let ts = chrono::NaiveDateTime::from_timestamp_millis(last_pulled as i64).expect("invalid timestamp");
                     info!("no broker updates for some time, last update: {:?}", ts);
                 }
             }

safekeeper/src/http/routes.rs

@@ -350,7 +350,6 @@ async fn record_safekeeper_info(mut request: Request<Body>) -> Result<Response<B
         backup_lsn: sk_info.backup_lsn.0,
         local_start_lsn: sk_info.local_start_lsn.0,
         availability_zone: None,
-        standby_horizon: sk_info.standby_horizon.0,
     };
 
     let tli = GlobalTimelines::get(ttid).map_err(ApiError::from)?;

safekeeper/src/lib.rs

@@ -31,8 +31,6 @@ pub mod safekeeper;
 pub mod send_wal;
 pub mod state;
 pub mod timeline;
-pub mod timeline_manager;
-pub mod timelines_set;
 pub mod wal_backup;
 pub mod wal_backup_partial;
 pub mod wal_service;

safekeeper/src/metrics.rs

@@ -11,9 +11,8 @@ use futures::Future;
 use metrics::{
     core::{AtomicU64, Collector, Desc, GenericCounter, GenericGaugeVec, Opts},
     proto::MetricFamily,
-    register_int_counter, register_int_counter_pair, register_int_counter_pair_vec,
-    register_int_counter_vec, Gauge, IntCounter, IntCounterPair, IntCounterPairVec, IntCounterVec,
-    IntGaugeVec,
+    register_int_counter, register_int_counter_pair_vec, register_int_counter_vec, Gauge,
+    IntCounter, IntCounterPairVec, IntCounterVec, IntGaugeVec,
 };
 use once_cell::sync::Lazy;
 
@@ -163,29 +162,6 @@ pub static PARTIAL_BACKUP_UPLOADED_BYTES: Lazy<IntCounter> = Lazy::new(|| {
     )
     .expect("Failed to register safekeeper_partial_backup_uploaded_bytes_total counter")
 });
-pub static MANAGER_ITERATIONS_TOTAL: Lazy<IntCounter> = Lazy::new(|| {
-    register_int_counter!(
-        "safekeeper_manager_iterations_total",
-        "Number of iterations of the timeline manager task"
-    )
-    .expect("Failed to register safekeeper_manager_iterations_total counter")
-});
-pub static MANAGER_ACTIVE_CHANGES: Lazy<IntCounter> = Lazy::new(|| {
-    register_int_counter!(
-        "safekeeper_manager_active_changes_total",
-        "Number of timeline active status changes in the timeline manager task"
-    )
-    .expect("Failed to register safekeeper_manager_active_changes_total counter")
-});
-pub static WAL_BACKUP_TASKS: Lazy<IntCounterPair> = Lazy::new(|| {
-    register_int_counter_pair!(
-        "safekeeper_wal_backup_tasks_started_total",
-        "Number of active WAL backup tasks",
-        "safekeeper_wal_backup_tasks_finished_total",
-        "Number of finished WAL backup tasks",
-    )
-    .expect("Failed to register safekeeper_wal_backup_tasks_finished_total counter")
-});
 
 pub const LABEL_UNKNOWN: &str = "unknown";
 
@@ -638,7 +614,8 @@ impl Collector for TimelineCollector {
         self.written_wal_seconds.reset();
         self.flushed_wal_seconds.reset();
 
-        let timelines_count = GlobalTimelines::get_all().len();
+        let timelines = GlobalTimelines::get_all();
+        let timelines_count = timelines.len();
         let mut active_timelines_count = 0;
 
         // Prometheus Collector is sync, and data is stored under async lock. To
@@ -769,9 +746,9 @@ impl Collector for TimelineCollector {
 
 async fn collect_timeline_metrics() -> Vec<FullTimelineInfo> {
     let mut res = vec![];
-    let active_timelines = GlobalTimelines::get_global_broker_active_set().get_all();
+    let timelines = GlobalTimelines::get_all();
 
-    for tli in active_timelines {
+    for tli in timelines {
         if let Some(info) = tli.info_for_metrics().await {
             res.push(info);
         }

safekeeper/src/receive_wal.rs

@@ -45,9 +45,6 @@ const DEFAULT_FEEDBACK_CAPACITY: usize = 8;
 pub struct WalReceivers {
     mutex: Mutex<WalReceiversShared>,
     pageserver_feedback_tx: tokio::sync::broadcast::Sender<PageserverFeedback>,
-
-    num_computes_tx: tokio::sync::watch::Sender<usize>,
-    num_computes_rx: tokio::sync::watch::Receiver<usize>,
 }
 
 /// Id under which walreceiver is registered in shmem.
@@ -58,21 +55,16 @@ impl WalReceivers {
         let (pageserver_feedback_tx, _) =
             tokio::sync::broadcast::channel(DEFAULT_FEEDBACK_CAPACITY);
 
-        let (num_computes_tx, num_computes_rx) = tokio::sync::watch::channel(0usize);
-
         Arc::new(WalReceivers {
             mutex: Mutex::new(WalReceiversShared { slots: Vec::new() }),
             pageserver_feedback_tx,
-            num_computes_tx,
-            num_computes_rx,
         })
     }
 
     /// Register new walreceiver. Returned guard provides access to the slot and
     /// automatically deregisters in Drop.
     pub fn register(self: &Arc<WalReceivers>, conn_id: Option<ConnectionId>) -> WalReceiverGuard {
-        let mut shared = self.mutex.lock();
-        let slots = &mut shared.slots;
+        let slots = &mut self.mutex.lock().slots;
         let walreceiver = WalReceiverState {
             conn_id,
             status: WalReceiverStatus::Voting,
@@ -86,9 +78,6 @@ impl WalReceivers {
             slots.push(Some(walreceiver));
             pos
         };
-
-        self.update_num(&shared);
-
         WalReceiverGuard {
             id: pos,
             walreceivers: self.clone(),
@@ -110,18 +99,7 @@ impl WalReceivers {
 
     /// Get number of walreceivers (compute connections).
     pub fn get_num(self: &Arc<WalReceivers>) -> usize {
-        self.mutex.lock().get_num()
-    }
-
-    /// Get channel for number of walreceivers.
-    pub fn get_num_rx(self: &Arc<WalReceivers>) -> tokio::sync::watch::Receiver<usize> {
-        self.num_computes_rx.clone()
-    }
-
-    /// Should get called after every update of slots.
-    fn update_num(self: &Arc<WalReceivers>, shared: &MutexGuard<WalReceiversShared>) {
-        let num = shared.get_num();
-        self.num_computes_tx.send_replace(num);
+        self.mutex.lock().slots.iter().flatten().count()
     }
 
     /// Get state of all walreceivers.
@@ -145,7 +123,6 @@ impl WalReceivers {
     fn unregister(self: &Arc<WalReceivers>, id: WalReceiverId) {
         let mut shared = self.mutex.lock();
         shared.slots[id] = None;
-        self.update_num(&shared);
     }
 
     /// Broadcast pageserver feedback to connected walproposers.
@@ -160,13 +137,6 @@ struct WalReceiversShared {
     slots: Vec<Option<WalReceiverState>>,
 }
 
-impl WalReceiversShared {
-    /// Get number of walreceivers (compute connections).
-    fn get_num(&self) -> usize {
-        self.slots.iter().flatten().count()
-    }
-}
-
 #[derive(Debug, Clone, Serialize, Deserialize)]
 pub struct WalReceiverState {
     /// None means it is recovery initiated by us (this safekeeper).
@@ -213,19 +183,9 @@ impl SafekeeperPostgresHandler {
         &mut self,
         pgb: &mut PostgresBackend<IO>,
     ) -> Result<(), QueryError> {
-        let mut tli: Option<Arc<Timeline>> = None;
-        if let Err(end) = self.handle_start_wal_push_guts(pgb, &mut tli).await {
+        if let Err(end) = self.handle_start_wal_push_guts(pgb).await {
             // Log the result and probably send it to the client, closing the stream.
-            let handle_end_fut = pgb.handle_copy_stream_end(end);
-            // If we managed to create the timeline, augment logging with current LSNs etc.
-            if let Some(tli) = tli {
-                let info = tli.get_safekeeper_info(&self.conf).await;
-                handle_end_fut
-                    .instrument(info_span!("", term=%info.term, last_log_term=%info.last_log_term, flush_lsn=%Lsn(info.flush_lsn), commit_lsn=%Lsn(info.commit_lsn)))
-                    .await;
-            } else {
-                handle_end_fut.await;
-            }
+            pgb.handle_copy_stream_end(end).await;
         }
         Ok(())
     }
@@ -233,7 +193,6 @@ impl SafekeeperPostgresHandler {
     pub async fn handle_start_wal_push_guts<IO: AsyncRead + AsyncWrite + Unpin>(
         &mut self,
         pgb: &mut PostgresBackend<IO>,
-        tli: &mut Option<Arc<Timeline>>,
     ) -> Result<(), CopyStreamHandlerEnd> {
         // Notify the libpq client that it's allowed to send `CopyData` messages
         pgb.write_message(&BeMessage::CopyBothResponse).await?;
@@ -263,17 +222,13 @@ impl SafekeeperPostgresHandler {
         // Read first message and create timeline if needed.
         let res = network_reader.read_first_message().await;
 
-        let network_res = if let Ok((timeline, next_msg)) = res {
+        let res = if let Ok((tli, next_msg)) = res {
             let pageserver_feedback_rx: tokio::sync::broadcast::Receiver<PageserverFeedback> =
-                timeline
-                    .get_walreceivers()
-                    .pageserver_feedback_tx
-                    .subscribe();
-            *tli = Some(timeline.clone());
+                tli.get_walreceivers().pageserver_feedback_tx.subscribe();
 
             tokio::select! {
                 // todo: add read|write .context to these errors
-                r = network_reader.run(msg_tx, msg_rx, reply_tx, timeline.clone(), next_msg) => r,
+                r = network_reader.run(msg_tx, msg_rx, reply_tx, tli.clone(), next_msg) => r,
                 r = network_write(pgb, reply_rx, pageserver_feedback_rx) => r,
             }
         } else {
@@ -289,13 +244,13 @@ impl SafekeeperPostgresHandler {
         match acceptor_handle {
             None => {
                 // failed even before spawning; read_network should have error
-                Err(network_res.expect_err("no error with WalAcceptor not spawn"))
+                Err(res.expect_err("no error with WalAcceptor not spawn"))
             }
             Some(handle) => {
                 let wal_acceptor_res = handle.await;
 
                 // If there was any network error, return it.
-                network_res?;
+                res?;
 
                 // Otherwise, WalAcceptor thread must have errored.
                 match wal_acceptor_res {
@@ -486,7 +441,14 @@ impl WalAcceptor {
     /// The main loop. Returns Ok(()) if either msg_rx or reply_tx got closed;
     /// it must mean that network thread terminated.
     async fn run(&mut self) -> anyhow::Result<()> {
+        // Register the connection and defer unregister.
+        // Order of the next two lines is important: we want first to remove our entry and then
+        // update status which depends on registered connections.
+        let _compute_conn_guard = ComputeConnectionGuard {
+            timeline: Arc::clone(&self.tli),
+        };
         let walreceiver_guard = self.tli.get_walreceivers().register(self.conn_id);
+        self.tli.update_status_notify().await?;
 
         // After this timestamp we will stop processing AppendRequests and send a response
         // to the walproposer. walproposer sends at least one AppendRequest per second,
@@ -552,3 +514,19 @@ impl WalAcceptor {
         }
     }
 }
+
+/// Calls update_status_notify in drop to update timeline status.
+struct ComputeConnectionGuard {
+    timeline: Arc<Timeline>,
+}
+
+impl Drop for ComputeConnectionGuard {
+    fn drop(&mut self) {
+        let tli = self.timeline.clone();
+        tokio::spawn(async move {
+            if let Err(e) = tli.update_status_notify().await {
+                error!("failed to update timeline status: {}", e);
+            }
+        });
+    }
+}

safekeeper/src/recovery.rs

@@ -37,11 +37,17 @@ use crate::{
 #[instrument(name = "recovery task", skip_all, fields(ttid = %tli.ttid))]
 pub async fn recovery_main(tli: Arc<Timeline>, conf: SafeKeeperConf) {
     info!("started");
+    let mut cancellation_rx = match tli.get_cancellation_rx() {
+        Ok(rx) => rx,
+        Err(_) => {
+            info!("timeline canceled during task start");
+            return;
+        }
+    };
 
-    let cancel = tli.cancel.clone();
     select! {
         _ = recovery_main_loop(tli, conf) => { unreachable!() }
-        _ = cancel.cancelled() => {
+        _ = cancellation_rx.changed() => {
             info!("stopped");
         }
     }

safekeeper/src/remove_wal.rs

@@ -7,18 +7,29 @@ use tracing::*;
 
 use crate::{GlobalTimelines, SafeKeeperConf};
 
-pub async fn task_main(_conf: SafeKeeperConf) -> anyhow::Result<()> {
+const ALLOW_INACTIVE_TIMELINES: bool = true;
+
+pub async fn task_main(conf: SafeKeeperConf) -> anyhow::Result<()> {
     let wal_removal_interval = Duration::from_millis(5000);
     loop {
         let now = tokio::time::Instant::now();
+        let mut active_timelines = 0;
+
         let tlis = GlobalTimelines::get_all();
         for tli in &tlis {
+            let is_active = tli.is_active().await;
+            if is_active {
+                active_timelines += 1;
+            }
+            if !ALLOW_INACTIVE_TIMELINES && !is_active {
+                continue;
+            }
             let ttid = tli.ttid;
             async {
                 if let Err(e) = tli.maybe_persist_control_file().await {
                     warn!("failed to persist control file: {e}");
                 }
-                if let Err(e) = tli.remove_old_wal().await {
+                if let Err(e) = tli.remove_old_wal(conf.wal_backup_enabled).await {
                     error!("failed to remove WAL: {}", e);
                 }
             }
@@ -31,8 +42,8 @@ pub async fn task_main(_conf: SafeKeeperConf) -> anyhow::Result<()> {
 
         if elapsed > wal_removal_interval {
             info!(
-                "WAL removal is too long, processed {} timelines in {:?}",
-                total_timelines, elapsed
+                "WAL removal is too long, processed {} active timelines ({} total) in {:?}",
+                active_timelines, total_timelines, elapsed
             );
         }
 

safekeeper/src/safekeeper.rs

@@ -827,10 +827,10 @@ where
 
     /// Persist control file if there is something to save and enough time
     /// passed after the last save.
-    pub async fn maybe_persist_inmem_control_file(&mut self) -> Result<bool> {
+    pub async fn maybe_persist_inmem_control_file(&mut self) -> Result<()> {
         const CF_SAVE_INTERVAL: Duration = Duration::from_secs(300);
         if self.state.pers.last_persist_at().elapsed() < CF_SAVE_INTERVAL {
-            return Ok(false);
+            return Ok(());
         }
         let need_persist = self.state.inmem.commit_lsn > self.state.commit_lsn
             || self.state.inmem.backup_lsn > self.state.backup_lsn
@@ -840,7 +840,7 @@ where
             self.state.flush().await?;
             trace!("saved control file: {CF_SAVE_INTERVAL:?} passed");
         }
-        Ok(need_persist)
+        Ok(())
     }
 
     /// Handle request to append WAL.

safekeeper/src/send_wal.rs

@@ -23,7 +23,7 @@ use utils::failpoint_support;
 use utils::id::TenantTimelineId;
 use utils::pageserver_feedback::PageserverFeedback;
 
-use std::cmp::{max, min};
+use std::cmp::min;
 use std::net::SocketAddr;
 use std::str;
 use std::sync::Arc;
@@ -85,17 +85,8 @@ impl StandbyReply {
 
 #[derive(Debug, Clone, Copy, Serialize, Deserialize)]
 pub struct StandbyFeedback {
-    pub reply: StandbyReply,
-    pub hs_feedback: HotStandbyFeedback,
-}
-
-impl StandbyFeedback {
-    pub fn empty() -> Self {
-        StandbyFeedback {
-            reply: StandbyReply::empty(),
-            hs_feedback: HotStandbyFeedback::empty(),
-        }
-    }
+    reply: StandbyReply,
+    hs_feedback: HotStandbyFeedback,
 }
 
 /// WalSenders registry. Timeline holds it (wrapped in Arc).
@@ -171,8 +162,8 @@ impl WalSenders {
     }
 
     /// Get aggregated hot standby feedback (we send it to compute).
-    pub fn get_hotstandby(self: &Arc<WalSenders>) -> StandbyFeedback {
-        self.mutex.lock().agg_standby_feedback
+    pub fn get_hotstandby(self: &Arc<WalSenders>) -> HotStandbyFeedback {
+        self.mutex.lock().agg_hs_feedback
     }
 
     /// Record new pageserver feedback, update aggregated values.
@@ -193,10 +184,6 @@ impl WalSenders {
     fn record_standby_reply(self: &Arc<WalSenders>, id: WalSenderId, reply: &StandbyReply) {
         let mut shared = self.mutex.lock();
         let slot = shared.get_slot_mut(id);
-        debug!(
-            "Record standby reply: ts={} apply_lsn={}",
-            reply.reply_ts, reply.apply_lsn
-        );
         match &mut slot.feedback {
             ReplicationFeedback::Standby(sf) => sf.reply = *reply,
             ReplicationFeedback::Pageserver(_) => {
@@ -221,7 +208,7 @@ impl WalSenders {
                 })
             }
         }
-        shared.update_reply_feedback();
+        shared.update_hs_feedback();
     }
 
     /// Get remote_consistent_lsn reported by the pageserver. Returns None if
@@ -239,13 +226,13 @@ impl WalSenders {
     fn unregister(self: &Arc<WalSenders>, id: WalSenderId) {
         let mut shared = self.mutex.lock();
         shared.slots[id] = None;
-        shared.update_reply_feedback();
+        shared.update_hs_feedback();
     }
 }
 
 struct WalSendersShared {
     // aggregated over all walsenders value
-    agg_standby_feedback: StandbyFeedback,
+    agg_hs_feedback: HotStandbyFeedback,
     // last feedback ever received from any pageserver, empty if none
     last_ps_feedback: PageserverFeedback,
     // total counter of pageserver feedbacks received
@@ -256,7 +243,7 @@ struct WalSendersShared {
 impl WalSendersShared {
     fn new() -> Self {
         WalSendersShared {
-            agg_standby_feedback: StandbyFeedback::empty(),
+            agg_hs_feedback: HotStandbyFeedback::empty(),
             last_ps_feedback: PageserverFeedback::empty(),
             ps_feedback_counter: 0,
             slots: Vec::new(),
@@ -273,11 +260,10 @@ impl WalSendersShared {
         self.slots[id].as_mut().expect("walsender doesn't exist")
     }
 
-    /// Update aggregated hot standy and normal reply feedbacks. We just take min of valid xmins
+    /// Update aggregated hot standy feedback. We just take min of valid xmins
     /// and ts.
-    fn update_reply_feedback(&mut self) {
+    fn update_hs_feedback(&mut self) {
         let mut agg = HotStandbyFeedback::empty();
-        let mut reply_agg = StandbyReply::empty();
         for ws_state in self.slots.iter().flatten() {
             if let ReplicationFeedback::Standby(standby_feedback) = ws_state.feedback {
                 let hs_feedback = standby_feedback.hs_feedback;
@@ -290,7 +276,7 @@ impl WalSendersShared {
                     } else {
                         agg.xmin = hs_feedback.xmin;
                     }
-                    agg.ts = max(agg.ts, hs_feedback.ts);
+                    agg.ts = min(agg.ts, hs_feedback.ts);
                 }
                 if hs_feedback.catalog_xmin != INVALID_FULL_TRANSACTION_ID {
                     if agg.catalog_xmin != INVALID_FULL_TRANSACTION_ID {
@@ -298,43 +284,11 @@ impl WalSendersShared {
                     } else {
                         agg.catalog_xmin = hs_feedback.catalog_xmin;
                     }
-                    agg.ts = max(agg.ts, hs_feedback.ts);
-                }
-                let reply = standby_feedback.reply;
-                if reply.write_lsn != Lsn::INVALID {
-                    if reply_agg.write_lsn != Lsn::INVALID {
-                        reply_agg.write_lsn = Lsn::min(reply_agg.write_lsn, reply.write_lsn);
-                    } else {
-                        reply_agg.write_lsn = reply.write_lsn;
-                    }
-                }
-                if reply.flush_lsn != Lsn::INVALID {
-                    if reply_agg.flush_lsn != Lsn::INVALID {
-                        reply_agg.flush_lsn = Lsn::min(reply_agg.flush_lsn, reply.flush_lsn);
-                    } else {
-                        reply_agg.flush_lsn = reply.flush_lsn;
-                    }
-                }
-                if reply.apply_lsn != Lsn::INVALID {
-                    if reply_agg.apply_lsn != Lsn::INVALID {
-                        reply_agg.apply_lsn = Lsn::min(reply_agg.apply_lsn, reply.apply_lsn);
-                    } else {
-                        reply_agg.apply_lsn = reply.apply_lsn;
-                    }
-                }
-                if reply.reply_ts != 0 {
-                    if reply_agg.reply_ts != 0 {
-                        reply_agg.reply_ts = TimestampTz::min(reply_agg.reply_ts, reply.reply_ts);
-                    } else {
-                        reply_agg.reply_ts = reply.reply_ts;
-                    }
+                    agg.ts = min(agg.ts, hs_feedback.ts);
                 }
             }
         }
-        self.agg_standby_feedback = StandbyFeedback {
-            reply: reply_agg,
-            hs_feedback: agg,
-        };
+        self.agg_hs_feedback = agg;
     }
 }
 
@@ -386,16 +340,12 @@ impl SafekeeperPostgresHandler {
         start_pos: Lsn,
         term: Option<Term>,
     ) -> Result<(), QueryError> {
-        let tli = GlobalTimelines::get(self.ttid).map_err(|e| QueryError::Other(e.into()))?;
         if let Err(end) = self
-            .handle_start_replication_guts(pgb, start_pos, term, tli.clone())
+            .handle_start_replication_guts(pgb, start_pos, term)
             .await
         {
-            let info = tli.get_safekeeper_info(&self.conf).await;
             // Log the result and probably send it to the client, closing the stream.
-            pgb.handle_copy_stream_end(end)
-            .instrument(info_span!("", term=%info.term, last_log_term=%info.last_log_term, flush_lsn=%Lsn(info.flush_lsn), commit_lsn=%Lsn(info.flush_lsn)))
-            .await;
+            pgb.handle_copy_stream_end(end).await;
         }
         Ok(())
     }
@@ -405,9 +355,10 @@ impl SafekeeperPostgresHandler {
         pgb: &mut PostgresBackend<IO>,
         start_pos: Lsn,
         term: Option<Term>,
-        tli: Arc<Timeline>,
     ) -> Result<(), CopyStreamHandlerEnd> {
         let appname = self.appname.clone();
+        let tli =
+            GlobalTimelines::get(self.ttid).map_err(|e| CopyStreamHandlerEnd::Other(e.into()))?;
 
         // Use a guard object to remove our entry from the timeline when we are done.
         let ws_guard = Arc::new(tli.get_walsenders().register(
@@ -756,15 +707,8 @@ impl<IO: AsyncRead + AsyncWrite + Unpin> ReplyReader<IO> {
         match msg.first().cloned() {
             Some(HOT_STANDBY_FEEDBACK_TAG_BYTE) => {
                 // Note: deserializing is on m[1..] because we skip the tag byte.
-                let mut hs_feedback = HotStandbyFeedback::des(&msg[1..])
+                let hs_feedback = HotStandbyFeedback::des(&msg[1..])
                     .context("failed to deserialize HotStandbyFeedback")?;
-                // TODO: xmin/catalog_xmin are serialized by walreceiver.c in this way:
-                // pq_sendint32(&reply_message, xmin);
-                // pq_sendint32(&reply_message, xmin_epoch);
-                // So it is two big endian 32-bit words in low endian order!
-                hs_feedback.xmin = (hs_feedback.xmin >> 32) | (hs_feedback.xmin << 32);
-                hs_feedback.catalog_xmin =
-                    (hs_feedback.catalog_xmin >> 32) | (hs_feedback.catalog_xmin << 32);
                 self.ws_guard
                     .walsenders
                     .record_hs_feedback(self.ws_guard.id, &hs_feedback);
@@ -846,11 +790,8 @@ mod tests {
     fn test_hs_feedback_no_valid() {
         let mut wss = WalSendersShared::new();
         push_feedback(&mut wss, hs_feedback(1, INVALID_FULL_TRANSACTION_ID));
-        wss.update_reply_feedback();
-        assert_eq!(
-            wss.agg_standby_feedback.hs_feedback.xmin,
-            INVALID_FULL_TRANSACTION_ID
-        );
+        wss.update_hs_feedback();
+        assert_eq!(wss.agg_hs_feedback.xmin, INVALID_FULL_TRANSACTION_ID);
     }
 
     #[test]
@@ -859,7 +800,7 @@ mod tests {
         push_feedback(&mut wss, hs_feedback(1, INVALID_FULL_TRANSACTION_ID));
         push_feedback(&mut wss, hs_feedback(1, 42));
         push_feedback(&mut wss, hs_feedback(1, 64));
-        wss.update_reply_feedback();
-        assert_eq!(wss.agg_standby_feedback.hs_feedback.xmin, 42);
+        wss.update_hs_feedback();
+        assert_eq!(wss.agg_hs_feedback.xmin, 42);
     }
 }

safekeeper/src/timeline.rs

@@ -6,15 +6,15 @@ use camino::Utf8PathBuf;
 use postgres_ffi::XLogSegNo;
 use serde::{Deserialize, Serialize};
 use tokio::fs;
-use tokio_util::sync::CancellationToken;
 
 use std::cmp::max;
-use std::ops::{Deref, DerefMut};
-use std::sync::atomic::{AtomicBool, Ordering};
 use std::sync::Arc;
 use std::time::Duration;
-use tokio::sync::{RwLock, RwLockReadGuard, RwLockWriteGuard};
-use tokio::{sync::watch, time::Instant};
+use tokio::sync::{Mutex, MutexGuard};
+use tokio::{
+    sync::{mpsc::Sender, watch},
+    time::Instant,
+};
 use tracing::*;
 use utils::http::error::ApiError;
 use utils::{
@@ -33,13 +33,12 @@ use crate::safekeeper::{
 };
 use crate::send_wal::WalSenders;
 use crate::state::{TimelineMemState, TimelinePersistentState};
-use crate::timelines_set::TimelinesSet;
 use crate::wal_backup::{self};
 use crate::{control_file, safekeeper::UNKNOWN_SERVER_VERSION};
 
 use crate::metrics::FullTimelineInfo;
 use crate::wal_storage::Storage as wal_storage_iface;
-use crate::{debug_dump, timeline_manager, wal_backup_partial, wal_storage};
+use crate::{debug_dump, wal_backup_partial, wal_storage};
 use crate::{GlobalTimelines, SafeKeeperConf};
 
 /// Things safekeeper should know about timeline state on peers.
@@ -52,7 +51,8 @@ pub struct PeerInfo {
     /// LSN of the last record.
     pub flush_lsn: Lsn,
     pub commit_lsn: Lsn,
-    /// Since which LSN safekeeper has WAL.
+    /// Since which LSN safekeeper has WAL. TODO: remove this once we fill new
+    /// sk since backup_lsn.
     pub local_start_lsn: Lsn,
     /// When info was received. Serde annotations are not very useful but make
     /// the code compile -- we don't rely on this field externally.
@@ -97,79 +97,25 @@ impl PeersInfo {
     }
 }
 
-pub type ReadGuardSharedState<'a> = RwLockReadGuard<'a, SharedState>;
-
-/// WriteGuardSharedState is a wrapper around `RwLockWriteGuard<SharedState>` that
-/// automatically updates `watch::Sender` channels with state on drop.
-pub struct WriteGuardSharedState<'a> {
-    tli: Arc<Timeline>,
-    guard: RwLockWriteGuard<'a, SharedState>,
-    skip_update: bool,
-}
-
-impl<'a> WriteGuardSharedState<'a> {
-    fn new(tli: Arc<Timeline>, guard: RwLockWriteGuard<'a, SharedState>) -> Self {
-        WriteGuardSharedState {
-            tli,
-            guard,
-            skip_update: false,
-        }
-    }
-}
-
-impl<'a> Deref for WriteGuardSharedState<'a> {
-    type Target = SharedState;
-
-    fn deref(&self) -> &Self::Target {
-        &self.guard
-    }
-}
-
-impl<'a> DerefMut for WriteGuardSharedState<'a> {
-    fn deref_mut(&mut self) -> &mut Self::Target {
-        &mut self.guard
-    }
-}
-
-impl<'a> Drop for WriteGuardSharedState<'a> {
-    fn drop(&mut self) {
-        let term_flush_lsn = TermLsn::from((self.guard.sk.get_term(), self.guard.sk.flush_lsn()));
-        let commit_lsn = self.guard.sk.state.inmem.commit_lsn;
-
-        let _ = self.tli.term_flush_lsn_watch_tx.send_if_modified(|old| {
-            if *old != term_flush_lsn {
-                *old = term_flush_lsn;
-                true
-            } else {
-                false
-            }
-        });
-
-        let _ = self.tli.commit_lsn_watch_tx.send_if_modified(|old| {
-            if *old != commit_lsn {
-                *old = commit_lsn;
-                true
-            } else {
-                false
-            }
-        });
-
-        if !self.skip_update {
-            // send notification about shared state update
-            self.tli.shared_state_version_tx.send_modify(|old| {
-                *old += 1;
-            });
-        }
-    }
-}
-
 /// Shared state associated with database instance
 pub struct SharedState {
     /// Safekeeper object
-    pub(crate) sk: SafeKeeper<control_file::FileStorage, wal_storage::PhysicalStorage>,
+    sk: SafeKeeper<control_file::FileStorage, wal_storage::PhysicalStorage>,
     /// In memory list containing state of peers sent in latest messages from them.
-    pub(crate) peers_info: PeersInfo,
-    pub(crate) last_removed_segno: XLogSegNo,
+    peers_info: PeersInfo,
+    /// True when WAL backup launcher oversees the timeline, making sure WAL is
+    /// offloaded, allows to bother launcher less.
+    wal_backup_active: bool,
+    /// True whenever there is at least some pending activity on timeline: live
+    /// compute connection, pageserver is not caughtup (it must have latest WAL
+    /// for new compute start) or WAL backuping is not finished. Practically it
+    /// means safekeepers broadcast info to peers about the timeline, old WAL is
+    /// trimmed.
+    ///
+    /// TODO: it might be better to remove tli completely from GlobalTimelines
+    /// when tli is inactive instead of having this flag.
+    active: bool,
+    last_removed_segno: XLogSegNo,
 }
 
 impl SharedState {
@@ -206,6 +152,8 @@ impl SharedState {
         Ok(Self {
             sk,
             peers_info: PeersInfo(vec![]),
+            wal_backup_active: false,
+            active: false,
             last_removed_segno: 0,
         })
     }
@@ -223,10 +171,75 @@ impl SharedState {
         Ok(Self {
             sk: SafeKeeper::new(control_store, wal_store, conf.my_id)?,
             peers_info: PeersInfo(vec![]),
+            wal_backup_active: false,
+            active: false,
             last_removed_segno: 0,
         })
     }
 
+    fn is_active(&self, num_computes: usize) -> bool {
+        self.is_wal_backup_required(num_computes)
+            // FIXME: add tracking of relevant pageservers and check them here individually,
+            // otherwise migration won't work (we suspend too early).
+            || self.sk.state.inmem.remote_consistent_lsn < self.sk.state.inmem.commit_lsn
+    }
+
+    /// Mark timeline active/inactive and return whether s3 offloading requires
+    /// start/stop action. If timeline is deactivated, control file is persisted
+    /// as maintenance task does that only for active timelines.
+    async fn update_status(&mut self, num_computes: usize, ttid: TenantTimelineId) -> bool {
+        let is_active = self.is_active(num_computes);
+        if self.active != is_active {
+            info!(
+                "timeline {} active={} now, remote_consistent_lsn={}, commit_lsn={}",
+                ttid,
+                is_active,
+                self.sk.state.inmem.remote_consistent_lsn,
+                self.sk.state.inmem.commit_lsn
+            );
+            if !is_active {
+                if let Err(e) = self.sk.state.flush().await {
+                    warn!("control file save in update_status failed: {:?}", e);
+                }
+            }
+        }
+        self.active = is_active;
+        self.is_wal_backup_action_pending(num_computes)
+    }
+
+    /// Should we run s3 offloading in current state?
+    fn is_wal_backup_required(&self, num_computes: usize) -> bool {
+        let seg_size = self.get_wal_seg_size();
+        num_computes > 0 ||
+        // Currently only the whole segment is offloaded, so compare segment numbers.
+            (self.sk.state.inmem.commit_lsn.segment_number(seg_size) >
+             self.sk.state.inmem.backup_lsn.segment_number(seg_size))
+    }
+
+    /// Is current state of s3 offloading is not what it ought to be?
+    fn is_wal_backup_action_pending(&self, num_computes: usize) -> bool {
+        let res = self.wal_backup_active != self.is_wal_backup_required(num_computes);
+        if res {
+            let action_pending = if self.is_wal_backup_required(num_computes) {
+                "start"
+            } else {
+                "stop"
+            };
+            trace!(
+                "timeline {} s3 offloading action {} pending: num_computes={}, commit_lsn={}, backup_lsn={}",
+                self.sk.state.timeline_id, action_pending, num_computes, self.sk.state.inmem.commit_lsn, self.sk.state.inmem.backup_lsn
+            );
+        }
+        res
+    }
+
+    /// Returns whether s3 offloading is required and sets current status as
+    /// matching.
+    fn wal_backup_attend(&mut self, num_computes: usize) -> bool {
+        self.wal_backup_active = self.is_wal_backup_required(num_computes);
+        self.wal_backup_active
+    }
+
     fn get_wal_seg_size(&self) -> usize {
         self.sk.state.server.wal_seg_size as usize
     }
@@ -235,7 +248,6 @@ impl SharedState {
         &self,
         ttid: &TenantTimelineId,
         conf: &SafeKeeperConf,
-        standby_apply_lsn: Lsn,
     ) -> SafekeeperTimelineInfo {
         SafekeeperTimelineInfo {
             safekeeper_id: conf.my_id.0,
@@ -258,14 +270,13 @@ impl SharedState {
             backup_lsn: self.sk.state.inmem.backup_lsn.0,
             local_start_lsn: self.sk.state.local_start_lsn.0,
             availability_zone: conf.availability_zone.clone(),
-            standby_horizon: standby_apply_lsn.0,
         }
     }
 
     /// Get our latest view of alive peers status on the timeline.
     /// We pass our own info through the broker as well, so when we don't have connection
     /// to the broker returned vec is empty.
-    pub(crate) fn get_peers(&self, heartbeat_timeout: Duration) -> Vec<PeerInfo> {
+    fn get_peers(&self, heartbeat_timeout: Duration) -> Vec<PeerInfo> {
         let now = Instant::now();
         self.peers_info
             .0
@@ -281,13 +292,18 @@ impl SharedState {
     /// offloading.
     /// While it is safe to use inmem values for determining horizon,
     /// we use persistent to make possible normal states less surprising.
-    fn get_horizon_segno(&self, extra_horizon_lsn: Option<Lsn>) -> XLogSegNo {
+    fn get_horizon_segno(
+        &self,
+        wal_backup_enabled: bool,
+        extra_horizon_lsn: Option<Lsn>,
+    ) -> XLogSegNo {
         let state = &self.sk.state;
 
         use std::cmp::min;
         let mut horizon_lsn = min(state.remote_consistent_lsn, state.peer_horizon_lsn);
-        // we don't want to remove WAL that is not yet offloaded to s3
-        horizon_lsn = min(horizon_lsn, state.backup_lsn);
+        if wal_backup_enabled {
+            horizon_lsn = min(horizon_lsn, state.backup_lsn);
+        }
         if let Some(extra_horizon_lsn) = extra_horizon_lsn {
             horizon_lsn = min(horizon_lsn, extra_horizon_lsn);
         }
@@ -328,6 +344,11 @@ impl From<TimelineError> for ApiError {
 pub struct Timeline {
     pub ttid: TenantTimelineId,
 
+    /// Sending here asks for wal backup launcher attention (start/stop
+    /// offloading). Sending ttid instead of concrete command allows to do
+    /// sending without timeline lock.
+    pub wal_backup_launcher_tx: Sender<TenantTimelineId>,
+
     /// Used to broadcast commit_lsn updates to all background jobs.
     commit_lsn_watch_tx: watch::Sender<Lsn>,
     commit_lsn_watch_rx: watch::Receiver<Lsn>,
@@ -339,19 +360,19 @@ pub struct Timeline {
     term_flush_lsn_watch_tx: watch::Sender<TermLsn>,
     term_flush_lsn_watch_rx: watch::Receiver<TermLsn>,
 
-    /// Broadcasts shared state updates.
-    shared_state_version_tx: watch::Sender<usize>,
-    shared_state_version_rx: watch::Receiver<usize>,
-
     /// Safekeeper and other state, that should remain consistent and
     /// synchronized with the disk. This is tokio mutex as we write WAL to disk
     /// while holding it, ensuring that consensus checks are in order.
-    mutex: RwLock<SharedState>,
+    mutex: Mutex<SharedState>,
     walsenders: Arc<WalSenders>,
     walreceivers: Arc<WalReceivers>,
 
-    /// Delete/cancel will trigger this, background tasks should drop out as soon as it fires
-    pub(crate) cancel: CancellationToken,
+    /// Cancellation channel. Delete/cancel will send `true` here as a cancellation signal.
+    cancellation_tx: watch::Sender<bool>,
+
+    /// Timeline should not be used after cancellation. Background tasks should
+    /// monitor this channel and stop eventually after receiving `true` from this channel.
+    cancellation_rx: watch::Receiver<bool>,
 
     /// Directory where timeline state is stored.
     pub timeline_dir: Utf8PathBuf,
@@ -361,15 +382,15 @@ pub struct Timeline {
     /// with different speed.
     // TODO: add `Arc<SafeKeeperConf>` here instead of adding each field separately.
     walsenders_keep_horizon: bool,
-
-    // timeline_manager controlled state
-    pub(crate) broker_active: AtomicBool,
-    pub(crate) wal_backup_active: AtomicBool,
 }
 
 impl Timeline {
     /// Load existing timeline from disk.
-    pub fn load_timeline(conf: &SafeKeeperConf, ttid: TenantTimelineId) -> Result<Timeline> {
+    pub fn load_timeline(
+        conf: &SafeKeeperConf,
+        ttid: TenantTimelineId,
+        wal_backup_launcher_tx: Sender<TenantTimelineId>,
+    ) -> Result<Timeline> {
         let _enter = info_span!("load_timeline", timeline = %ttid.timeline_id).entered();
 
         let shared_state = SharedState::restore(conf, &ttid)?;
@@ -379,25 +400,23 @@ impl Timeline {
             shared_state.sk.get_term(),
             shared_state.sk.flush_lsn(),
         )));
-        let (shared_state_version_tx, shared_state_version_rx) = watch::channel(0);
+        let (cancellation_tx, cancellation_rx) = watch::channel(false);
 
         let walreceivers = WalReceivers::new();
         Ok(Timeline {
             ttid,
+            wal_backup_launcher_tx,
             commit_lsn_watch_tx,
             commit_lsn_watch_rx,
             term_flush_lsn_watch_tx,
             term_flush_lsn_watch_rx,
-            shared_state_version_tx,
-            shared_state_version_rx,
-            mutex: RwLock::new(shared_state),
+            mutex: Mutex::new(shared_state),
             walsenders: WalSenders::new(walreceivers.clone()),
             walreceivers,
-            cancel: CancellationToken::default(),
+            cancellation_rx,
+            cancellation_tx,
             timeline_dir: conf.timeline_dir(&ttid),
             walsenders_keep_horizon: conf.walsenders_keep_horizon,
-            broker_active: AtomicBool::new(false),
-            wal_backup_active: AtomicBool::new(false),
         })
     }
 
@@ -405,6 +424,7 @@ impl Timeline {
     pub fn create_empty(
         conf: &SafeKeeperConf,
         ttid: TenantTimelineId,
+        wal_backup_launcher_tx: Sender<TenantTimelineId>,
         server_info: ServerInfo,
         commit_lsn: Lsn,
         local_start_lsn: Lsn,
@@ -412,28 +432,25 @@ impl Timeline {
         let (commit_lsn_watch_tx, commit_lsn_watch_rx) = watch::channel(Lsn::INVALID);
         let (term_flush_lsn_watch_tx, term_flush_lsn_watch_rx) =
             watch::channel(TermLsn::from((INVALID_TERM, Lsn::INVALID)));
-        let (shared_state_version_tx, shared_state_version_rx) = watch::channel(0);
-
+        let (cancellation_tx, cancellation_rx) = watch::channel(false);
         let state =
             TimelinePersistentState::new(&ttid, server_info, vec![], commit_lsn, local_start_lsn);
 
         let walreceivers = WalReceivers::new();
         Ok(Timeline {
             ttid,
+            wal_backup_launcher_tx,
             commit_lsn_watch_tx,
             commit_lsn_watch_rx,
             term_flush_lsn_watch_tx,
             term_flush_lsn_watch_rx,
-            shared_state_version_tx,
-            shared_state_version_rx,
-            mutex: RwLock::new(SharedState::create_new(conf, &ttid, state)?),
+            mutex: Mutex::new(SharedState::create_new(conf, &ttid, state)?),
             walsenders: WalSenders::new(walreceivers.clone()),
             walreceivers,
-            cancel: CancellationToken::default(),
+            cancellation_rx,
+            cancellation_tx,
             timeline_dir: conf.timeline_dir(&ttid),
             walsenders_keep_horizon: conf.walsenders_keep_horizon,
-            broker_active: AtomicBool::new(false),
-            wal_backup_active: AtomicBool::new(false),
         })
     }
 
@@ -444,9 +461,8 @@ impl Timeline {
     /// and state on disk should remain unchanged.
     pub async fn init_new(
         self: &Arc<Timeline>,
-        shared_state: &mut WriteGuardSharedState<'_>,
+        shared_state: &mut MutexGuard<'_, SharedState>,
         conf: &SafeKeeperConf,
-        broker_active_set: Arc<TimelinesSet>,
     ) -> Result<()> {
         match fs::metadata(&self.timeline_dir).await {
             Ok(_) => {
@@ -477,29 +493,16 @@ impl Timeline {
 
             return Err(e);
         }
-        self.bootstrap(conf, broker_active_set);
+        self.bootstrap(conf);
         Ok(())
     }
 
-    /// Bootstrap new or existing timeline starting background tasks.
-    pub fn bootstrap(
-        self: &Arc<Timeline>,
-        conf: &SafeKeeperConf,
-        broker_active_set: Arc<TimelinesSet>,
-    ) {
-        // Start manager task which will monitor timeline state and update
-        // background tasks.
-        tokio::spawn(timeline_manager::main_task(
-            self.clone(),
-            conf.clone(),
-            broker_active_set,
-        ));
-
+    /// Bootstrap new or existing timeline starting background stasks.
+    pub fn bootstrap(self: &Arc<Timeline>, conf: &SafeKeeperConf) {
         // Start recovery task which always runs on the timeline.
         if conf.peer_recovery_enabled {
             tokio::spawn(recovery_main(self.clone(), conf.clone()));
         }
-        // TODO: migrate to timeline_manager
         if conf.is_wal_backup_enabled() && conf.partial_backup_enabled {
             tokio::spawn(wal_backup_partial::main_task(self.clone(), conf.clone()));
         }
@@ -512,9 +515,10 @@ impl Timeline {
     /// deletion API endpoint is retriable.
     pub async fn delete(
         &self,
-        shared_state: &mut WriteGuardSharedState<'_>,
+        shared_state: &mut MutexGuard<'_, SharedState>,
         only_local: bool,
-    ) -> Result<bool> {
+    ) -> Result<(bool, bool)> {
+        let was_active = shared_state.active;
         self.cancel(shared_state);
 
         // TODO: It's better to wait for s3 offloader termination before
@@ -528,14 +532,20 @@ impl Timeline {
             wal_backup::delete_timeline(&self.ttid).await?;
         }
         let dir_existed = delete_dir(&self.timeline_dir).await?;
-        Ok(dir_existed)
+        Ok((dir_existed, was_active))
     }
 
     /// Cancel timeline to prevent further usage. Background tasks will stop
     /// eventually after receiving cancellation signal.
-    fn cancel(&self, shared_state: &mut WriteGuardSharedState<'_>) {
+    ///
+    /// Note that we can't notify backup launcher here while holding
+    /// shared_state lock, as this is a potential deadlock: caller is
+    /// responsible for that. Generally we should probably make WAL backup tasks
+    /// to shut down on their own, checking once in a while whether it is the
+    /// time.
+    fn cancel(&self, shared_state: &mut MutexGuard<'_, SharedState>) {
         info!("timeline {} is cancelled", self.ttid);
-        self.cancel.cancel();
+        let _ = self.cancellation_tx.send(true);
         // Close associated FDs. Nobody will be able to touch timeline data once
         // it is cancelled, so WAL storage won't be opened again.
         shared_state.sk.wal_store.close();
@@ -543,16 +553,44 @@ impl Timeline {
 
     /// Returns if timeline is cancelled.
     pub fn is_cancelled(&self) -> bool {
-        self.cancel.is_cancelled()
+        *self.cancellation_rx.borrow()
+    }
+
+    /// Returns watch channel which gets value when timeline is cancelled. It is
+    /// guaranteed to have not cancelled value observed (errors otherwise).
+    pub fn get_cancellation_rx(&self) -> Result<watch::Receiver<bool>> {
+        let rx = self.cancellation_rx.clone();
+        if *rx.borrow() {
+            bail!(TimelineError::Cancelled(self.ttid));
+        }
+        Ok(rx)
     }
 
     /// Take a writing mutual exclusive lock on timeline shared_state.
-    pub async fn write_shared_state<'a>(self: &'a Arc<Self>) -> WriteGuardSharedState<'a> {
-        WriteGuardSharedState::new(self.clone(), self.mutex.write().await)
+    pub async fn write_shared_state(&self) -> MutexGuard<SharedState> {
+        self.mutex.lock().await
     }
 
-    pub async fn read_shared_state(&self) -> ReadGuardSharedState {
-        self.mutex.read().await
+    async fn update_status(&self, shared_state: &mut SharedState) -> bool {
+        shared_state
+            .update_status(self.walreceivers.get_num(), self.ttid)
+            .await
+    }
+
+    /// Update timeline status and kick wal backup launcher to stop/start offloading if needed.
+    pub async fn update_status_notify(&self) -> Result<()> {
+        if self.is_cancelled() {
+            bail!(TimelineError::Cancelled(self.ttid));
+        }
+        let is_wal_backup_action_pending: bool = {
+            let mut shared_state = self.write_shared_state().await;
+            self.update_status(&mut shared_state).await
+        };
+        if is_wal_backup_action_pending {
+            // Can fail only if channel to a static thread got closed, which is not normal at all.
+            self.wal_backup_launcher_tx.send(self.ttid).await?;
+        }
+        Ok(())
     }
 
     /// Returns true if walsender should stop sending WAL to pageserver. We
@@ -564,7 +602,7 @@ impl Timeline {
         if self.is_cancelled() {
             return true;
         }
-        let shared_state = self.read_shared_state().await;
+        let shared_state = self.write_shared_state().await;
         if self.walreceivers.get_num() == 0 {
             return shared_state.sk.state.inmem.commit_lsn == Lsn(0) || // no data at all yet
             reported_remote_consistent_lsn >= shared_state.sk.state.inmem.commit_lsn;
@@ -572,9 +610,9 @@ impl Timeline {
         false
     }
 
-    /// Ensure that current term is t, erroring otherwise, and lock the state.
-    pub async fn acquire_term(&self, t: Term) -> Result<ReadGuardSharedState> {
-        let ss = self.read_shared_state().await;
+    /// Ensure taht current term is t, erroring otherwise, and lock the state.
+    pub async fn acquire_term(&self, t: Term) -> Result<MutexGuard<SharedState>> {
+        let ss = self.write_shared_state().await;
         if ss.sk.state.acceptor_state.term != t {
             bail!(
                 "failed to acquire term {}, current term {}",
@@ -585,6 +623,18 @@ impl Timeline {
         Ok(ss)
     }
 
+    /// Returns whether s3 offloading is required and sets current status as
+    /// matching it.
+    pub async fn wal_backup_attend(&self) -> bool {
+        if self.is_cancelled() {
+            return false;
+        }
+
+        self.write_shared_state()
+            .await
+            .wal_backup_attend(self.walreceivers.get_num())
+    }
+
     /// Returns commit_lsn watch channel.
     pub fn get_commit_lsn_watch_rx(&self) -> watch::Receiver<Lsn> {
         self.commit_lsn_watch_rx.clone()
@@ -595,14 +645,9 @@ impl Timeline {
         self.term_flush_lsn_watch_rx.clone()
     }
 
-    /// Returns watch channel for SharedState update version.
-    pub fn get_state_version_rx(&self) -> watch::Receiver<usize> {
-        self.shared_state_version_rx.clone()
-    }
-
     /// Pass arrived message to the safekeeper.
     pub async fn process_msg(
-        self: &Arc<Self>,
+        &self,
         msg: &ProposerAcceptorMessage,
     ) -> Result<Option<AcceptorProposerMessage>> {
         if self.is_cancelled() {
@@ -610,36 +655,53 @@ impl Timeline {
         }
 
         let mut rmsg: Option<AcceptorProposerMessage>;
+        let commit_lsn: Lsn;
+        let term_flush_lsn: TermLsn;
         {
             let mut shared_state = self.write_shared_state().await;
             rmsg = shared_state.sk.process_msg(msg).await?;
 
             // if this is AppendResponse, fill in proper hot standby feedback.
             if let Some(AcceptorProposerMessage::AppendResponse(ref mut resp)) = rmsg {
-                resp.hs_feedback = self.walsenders.get_hotstandby().hs_feedback;
+                resp.hs_feedback = self.walsenders.get_hotstandby();
             }
+
+            commit_lsn = shared_state.sk.state.inmem.commit_lsn;
+            term_flush_lsn =
+                TermLsn::from((shared_state.sk.get_term(), shared_state.sk.flush_lsn()));
         }
+        self.term_flush_lsn_watch_tx.send(term_flush_lsn)?;
+        self.commit_lsn_watch_tx.send(commit_lsn)?;
         Ok(rmsg)
     }
 
     /// Returns wal_seg_size.
     pub async fn get_wal_seg_size(&self) -> usize {
-        self.read_shared_state().await.get_wal_seg_size()
+        self.write_shared_state().await.get_wal_seg_size()
+    }
+
+    /// Returns true only if the timeline is loaded and active.
+    pub async fn is_active(&self) -> bool {
+        if self.is_cancelled() {
+            return false;
+        }
+
+        self.write_shared_state().await.active
     }
 
     /// Returns state of the timeline.
     pub async fn get_state(&self) -> (TimelineMemState, TimelinePersistentState) {
-        let state = self.read_shared_state().await;
+        let state = self.write_shared_state().await;
         (state.sk.state.inmem.clone(), state.sk.state.clone())
     }
 
     /// Returns latest backup_lsn.
     pub async fn get_wal_backup_lsn(&self) -> Lsn {
-        self.read_shared_state().await.sk.state.inmem.backup_lsn
+        self.write_shared_state().await.sk.state.inmem.backup_lsn
     }
 
     /// Sets backup_lsn to the given value.
-    pub async fn set_wal_backup_lsn(self: &Arc<Self>, backup_lsn: Lsn) -> Result<()> {
+    pub async fn set_wal_backup_lsn(&self, backup_lsn: Lsn) -> Result<()> {
         if self.is_cancelled() {
             bail!(TimelineError::Cancelled(self.ttid));
         }
@@ -653,34 +715,39 @@ impl Timeline {
 
     /// Get safekeeper info for broadcasting to broker and other peers.
     pub async fn get_safekeeper_info(&self, conf: &SafeKeeperConf) -> SafekeeperTimelineInfo {
-        let standby_apply_lsn = self.walsenders.get_hotstandby().reply.apply_lsn;
-        let shared_state = self.read_shared_state().await;
-        shared_state.get_safekeeper_info(&self.ttid, conf, standby_apply_lsn)
+        let shared_state = self.write_shared_state().await;
+        shared_state.get_safekeeper_info(&self.ttid, conf)
     }
 
     /// Update timeline state with peer safekeeper data.
-    pub async fn record_safekeeper_info(
-        self: &Arc<Self>,
-        sk_info: SafekeeperTimelineInfo,
-    ) -> Result<()> {
+    pub async fn record_safekeeper_info(&self, sk_info: SafekeeperTimelineInfo) -> Result<()> {
+        let is_wal_backup_action_pending: bool;
+        let commit_lsn: Lsn;
         {
             let mut shared_state = self.write_shared_state().await;
             shared_state.sk.record_safekeeper_info(&sk_info).await?;
             let peer_info = PeerInfo::from_sk_info(&sk_info, Instant::now());
             shared_state.peers_info.upsert(&peer_info);
+            is_wal_backup_action_pending = self.update_status(&mut shared_state).await;
+            commit_lsn = shared_state.sk.state.inmem.commit_lsn;
+        }
+        self.commit_lsn_watch_tx.send(commit_lsn)?;
+        // Wake up wal backup launcher, if it is time to stop the offloading.
+        if is_wal_backup_action_pending {
+            self.wal_backup_launcher_tx.send(self.ttid).await?;
         }
         Ok(())
     }
 
     /// Update in memory remote consistent lsn.
-    pub async fn update_remote_consistent_lsn(self: &Arc<Self>, candidate: Lsn) {
+    pub async fn update_remote_consistent_lsn(&self, candidate: Lsn) {
         let mut shared_state = self.write_shared_state().await;
         shared_state.sk.state.inmem.remote_consistent_lsn =
             max(shared_state.sk.state.inmem.remote_consistent_lsn, candidate);
     }
 
     pub async fn get_peers(&self, conf: &SafeKeeperConf) -> Vec<PeerInfo> {
-        let shared_state = self.read_shared_state().await;
+        let shared_state = self.write_shared_state().await;
         shared_state.get_peers(conf.heartbeat_timeout)
     }
 
@@ -702,7 +769,7 @@ impl Timeline {
     /// depending on assembled quorum (e.g. classic picture 8 from Raft paper).
     /// Thus we don't try to predict it here.
     pub async fn recovery_needed(&self, heartbeat_timeout: Duration) -> RecoveryNeededInfo {
-        let ss = self.read_shared_state().await;
+        let ss = self.write_shared_state().await;
         let term = ss.sk.state.acceptor_state.term;
         let last_log_term = ss.sk.get_epoch();
         let flush_lsn = ss.sk.flush_lsn();
@@ -773,12 +840,12 @@ impl Timeline {
 
     /// Returns flush_lsn.
     pub async fn get_flush_lsn(&self) -> Lsn {
-        self.read_shared_state().await.sk.wal_store.flush_lsn()
+        self.write_shared_state().await.sk.wal_store.flush_lsn()
     }
 
     /// Delete WAL segments from disk that are no longer needed. This is determined
     /// based on pageserver's remote_consistent_lsn and local backup_lsn/peer_lsn.
-    pub async fn remove_old_wal(self: &Arc<Self>) -> Result<()> {
+    pub async fn remove_old_wal(&self, wal_backup_enabled: bool) -> Result<()> {
         if self.is_cancelled() {
             bail!(TimelineError::Cancelled(self.ttid));
         }
@@ -794,8 +861,9 @@ impl Timeline {
 
         let horizon_segno: XLogSegNo;
         let remover = {
-            let shared_state = self.read_shared_state().await;
-            horizon_segno = shared_state.get_horizon_segno(replication_horizon_lsn);
+            let shared_state = self.write_shared_state().await;
+            horizon_segno =
+                shared_state.get_horizon_segno(wal_backup_enabled, replication_horizon_lsn);
             if horizon_segno <= 1 || horizon_segno <= shared_state.last_removed_segno {
                 return Ok(()); // nothing to do
             }
@@ -809,11 +877,7 @@ impl Timeline {
 
         // update last_removed_segno
         let mut shared_state = self.write_shared_state().await;
-        if shared_state.last_removed_segno != horizon_segno {
-            shared_state.last_removed_segno = horizon_segno;
-        } else {
-            shared_state.skip_update = true;
-        }
+        shared_state.last_removed_segno = horizon_segno;
         Ok(())
     }
 
@@ -821,40 +885,46 @@ impl Timeline {
     /// passed after the last save. This helps to keep remote_consistent_lsn up
     /// to date so that storage nodes restart doesn't cause many pageserver ->
     /// safekeeper reconnections.
-    pub async fn maybe_persist_control_file(self: &Arc<Self>) -> Result<()> {
-        let mut guard = self.write_shared_state().await;
-        let changed = guard.sk.maybe_persist_inmem_control_file().await?;
-        guard.skip_update = !changed;
-        Ok(())
+    pub async fn maybe_persist_control_file(&self) -> Result<()> {
+        self.write_shared_state()
+            .await
+            .sk
+            .maybe_persist_inmem_control_file()
+            .await
     }
 
-    /// Gather timeline data for metrics.
+    /// Gather timeline data for metrics. If the timeline is not active, returns
+    /// None, we do not collect these.
     pub async fn info_for_metrics(&self) -> Option<FullTimelineInfo> {
         if self.is_cancelled() {
             return None;
         }
 
         let (ps_feedback_count, last_ps_feedback) = self.walsenders.get_ps_feedback_stats();
-        let state = self.read_shared_state().await;
-        Some(FullTimelineInfo {
-            ttid: self.ttid,
-            ps_feedback_count,
-            last_ps_feedback,
-            wal_backup_active: self.wal_backup_active.load(Ordering::Relaxed),
-            timeline_is_active: self.broker_active.load(Ordering::Relaxed),
-            num_computes: self.walreceivers.get_num() as u32,
-            last_removed_segno: state.last_removed_segno,
-            epoch_start_lsn: state.sk.epoch_start_lsn,
-            mem_state: state.sk.state.inmem.clone(),
-            persisted_state: state.sk.state.clone(),
-            flush_lsn: state.sk.wal_store.flush_lsn(),
-            wal_storage: state.sk.wal_store.get_metrics(),
-        })
+        let state = self.write_shared_state().await;
+        if state.active {
+            Some(FullTimelineInfo {
+                ttid: self.ttid,
+                ps_feedback_count,
+                last_ps_feedback,
+                wal_backup_active: state.wal_backup_active,
+                timeline_is_active: state.active,
+                num_computes: self.walreceivers.get_num() as u32,
+                last_removed_segno: state.last_removed_segno,
+                epoch_start_lsn: state.sk.epoch_start_lsn,
+                mem_state: state.sk.state.inmem.clone(),
+                persisted_state: state.sk.state.clone(),
+                flush_lsn: state.sk.wal_store.flush_lsn(),
+                wal_storage: state.sk.wal_store.get_metrics(),
+            })
+        } else {
+            None
+        }
     }
 
     /// Returns in-memory timeline state to build a full debug dump.
     pub async fn memory_dump(&self) -> debug_dump::Memory {
-        let state = self.read_shared_state().await;
+        let state = self.write_shared_state().await;
 
         let (write_lsn, write_record_lsn, flush_lsn, file_open) =
             state.sk.wal_store.internal_state();
@@ -863,8 +933,8 @@ impl Timeline {
             is_cancelled: self.is_cancelled(),
             peers_info_len: state.peers_info.0.len(),
             walsenders: self.walsenders.get_all(),
-            wal_backup_active: self.wal_backup_active.load(Ordering::Relaxed),
-            active: self.broker_active.load(Ordering::Relaxed),
+            wal_backup_active: state.wal_backup_active,
+            active: state.active,
             num_computes: self.walreceivers.get_num() as u32,
             last_removed_segno: state.last_removed_segno,
             epoch_start_lsn: state.sk.epoch_start_lsn,
@@ -878,7 +948,7 @@ impl Timeline {
 
     /// Apply a function to the control file state and persist it.
     pub async fn map_control_file<T>(
-        self: &Arc<Self>,
+        &self,
         f: impl FnOnce(&mut TimelinePersistentState) -> Result<T>,
     ) -> Result<T> {
         let mut state = self.write_shared_state().await;

safekeeper/src/timeline_manager.rs

@@ -1,145 +0,0 @@
-//! The timeline manager task is responsible for managing the timeline's background tasks.
-//! It is spawned alongside each timeline and exits when the timeline is deleted.
-//! It watches for changes in the timeline state and decides when to spawn or kill background tasks.
-//! It also can manage some reactive state, like should the timeline be active for broker pushes or not.
-
-use std::{sync::Arc, time::Duration};
-
-use tracing::{info, instrument, warn};
-use utils::lsn::Lsn;
-
-use crate::{
-    metrics::{MANAGER_ACTIVE_CHANGES, MANAGER_ITERATIONS_TOTAL},
-    timeline::{PeerInfo, ReadGuardSharedState, Timeline},
-    timelines_set::TimelinesSet,
-    wal_backup::{self, WalBackupTaskHandle},
-    SafeKeeperConf,
-};
-
-pub struct StateSnapshot {
-    pub commit_lsn: Lsn,
-    pub backup_lsn: Lsn,
-    pub remote_consistent_lsn: Lsn,
-    pub peers: Vec<PeerInfo>,
-}
-
-impl StateSnapshot {
-    /// Create a new snapshot of the timeline state.
-    fn new(read_guard: ReadGuardSharedState, heartbeat_timeout: Duration) -> Self {
-        Self {
-            commit_lsn: read_guard.sk.state.inmem.commit_lsn,
-            backup_lsn: read_guard.sk.state.inmem.backup_lsn,
-            remote_consistent_lsn: read_guard.sk.state.inmem.remote_consistent_lsn,
-            peers: read_guard.get_peers(heartbeat_timeout),
-        }
-    }
-}
-
-/// Control how often the manager task should wake up to check updates.
-/// There is no need to check for updates more often than this.
-const REFRESH_INTERVAL: Duration = Duration::from_millis(300);
-
-/// This task gets spawned alongside each timeline and is responsible for managing the timeline's
-/// background tasks.
-#[instrument(name = "manager", skip_all, fields(ttid = %tli.ttid))]
-pub async fn main_task(
-    tli: Arc<Timeline>,
-    conf: SafeKeeperConf,
-    broker_active_set: Arc<TimelinesSet>,
-) {
-    scopeguard::defer! {
-        if tli.is_cancelled() {
-            info!("manager task finished");
-        } else {
-            warn!("manager task finished prematurely");
-        }
-    };
-
-    // sets whether timeline is active for broker pushes or not
-    let mut tli_broker_active = broker_active_set.guard(tli.clone());
-
-    let ttid = tli.ttid;
-    let wal_seg_size = tli.get_wal_seg_size().await;
-    let heartbeat_timeout = conf.heartbeat_timeout;
-
-    let mut state_version_rx = tli.get_state_version_rx();
-
-    let walreceivers = tli.get_walreceivers();
-    let mut num_computes_rx = walreceivers.get_num_rx();
-
-    // list of background tasks
-    let mut backup_task: Option<WalBackupTaskHandle> = None;
-
-    let last_state = 'outer: loop {
-        MANAGER_ITERATIONS_TOTAL.inc();
-
-        let state_snapshot = StateSnapshot::new(tli.read_shared_state().await, heartbeat_timeout);
-        let num_computes = *num_computes_rx.borrow();
-
-        let is_wal_backup_required =
-            wal_backup::is_wal_backup_required(wal_seg_size, num_computes, &state_snapshot);
-
-        if conf.is_wal_backup_enabled() {
-            wal_backup::update_task(
-                &conf,
-                ttid,
-                is_wal_backup_required,
-                &state_snapshot,
-                &mut backup_task,
-            )
-            .await;
-        }
-
-        let is_active = is_wal_backup_required
-            || num_computes > 0
-            || state_snapshot.remote_consistent_lsn < state_snapshot.commit_lsn;
-
-        // update the broker timeline set
-        if tli_broker_active.set(is_active) {
-            // write log if state has changed
-            info!(
-                "timeline active={} now, remote_consistent_lsn={}, commit_lsn={}",
-                is_active, state_snapshot.remote_consistent_lsn, state_snapshot.commit_lsn,
-            );
-
-            MANAGER_ACTIVE_CHANGES.inc();
-
-            if !is_active {
-                // TODO: maybe use tokio::spawn?
-                if let Err(e) = tli.maybe_persist_control_file().await {
-                    warn!("control file save in update_status failed: {:?}", e);
-                }
-            }
-        }
-
-        // update the state in Arc<Timeline>
-        tli.wal_backup_active
-            .store(backup_task.is_some(), std::sync::atomic::Ordering::Relaxed);
-        tli.broker_active
-            .store(is_active, std::sync::atomic::Ordering::Relaxed);
-
-        // wait until something changes. tx channels are stored under Arc, so they will not be
-        // dropped until the manager task is finished.
-        tokio::select! {
-            _ = tli.cancel.cancelled() => {
-                // timeline was deleted
-                break 'outer state_snapshot;
-            }
-            _ = async {
-                // don't wake up on every state change, but at most every REFRESH_INTERVAL
-                tokio::time::sleep(REFRESH_INTERVAL).await;
-                let _ = state_version_rx.changed().await;
-            } => {
-                // state was updated
-            }
-            _ = num_computes_rx.changed() => {
-                // number of connected computes was updated
-            }
-        }
-    };
-
-    // shutdown background tasks
-    if conf.is_wal_backup_enabled() {
-        wal_backup::update_task(&conf, ttid, false, &last_state, &mut backup_task).await;
-    }
-}